diff options
Diffstat (limited to 'gl/regexec.c')
| -rw-r--r-- | gl/regexec.c | 4398 |
1 files changed, 4398 insertions, 0 deletions
diff --git a/gl/regexec.c b/gl/regexec.c new file mode 100644 index 00000000..7c186aa2 --- /dev/null +++ b/gl/regexec.c | |||
| @@ -0,0 +1,4398 @@ | |||
| 1 | /* Extended regular expression matching and search library. | ||
| 2 | Copyright (C) 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc. | ||
| 3 | This file is part of the GNU C Library. | ||
| 4 | Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>. | ||
| 5 | |||
| 6 | This program is free software; you can redistribute it and/or modify | ||
| 7 | it under the terms of the GNU General Public License as published by | ||
| 8 | the Free Software Foundation; either version 2, or (at your option) | ||
| 9 | any later version. | ||
| 10 | |||
| 11 | This program is distributed in the hope that it will be useful, | ||
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 14 | GNU General Public License for more details. | ||
| 15 | |||
| 16 | You should have received a copy of the GNU General Public License along | ||
| 17 | with this program; if not, write to the Free Software Foundation, | ||
| 18 | Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */ | ||
| 19 | |||
| 20 | static reg_errcode_t match_ctx_init (re_match_context_t *cache, int eflags, | ||
| 21 | Idx n) internal_function; | ||
| 22 | static void match_ctx_clean (re_match_context_t *mctx) internal_function; | ||
| 23 | static void match_ctx_free (re_match_context_t *cache) internal_function; | ||
| 24 | static reg_errcode_t match_ctx_add_entry (re_match_context_t *cache, Idx node, | ||
| 25 | Idx str_idx, Idx from, Idx to) | ||
| 26 | internal_function; | ||
| 27 | static Idx search_cur_bkref_entry (const re_match_context_t *mctx, Idx str_idx) | ||
| 28 | internal_function; | ||
| 29 | static reg_errcode_t match_ctx_add_subtop (re_match_context_t *mctx, Idx node, | ||
| 30 | Idx str_idx) internal_function; | ||
| 31 | static re_sub_match_last_t * match_ctx_add_sublast (re_sub_match_top_t *subtop, | ||
| 32 | Idx node, Idx str_idx) | ||
| 33 | internal_function; | ||
| 34 | static void sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts, | ||
| 35 | re_dfastate_t **limited_sts, Idx last_node, | ||
| 36 | Idx last_str_idx) | ||
| 37 | internal_function; | ||
| 38 | static reg_errcode_t re_search_internal (const regex_t *preg, | ||
| 39 | const char *string, Idx length, | ||
| 40 | Idx start, Idx last_start, Idx stop, | ||
| 41 | size_t nmatch, regmatch_t pmatch[], | ||
| 42 | int eflags) internal_function; | ||
| 43 | static regoff_t re_search_2_stub (struct re_pattern_buffer *bufp, | ||
| 44 | const char *string1, Idx length1, | ||
| 45 | const char *string2, Idx length2, | ||
| 46 | Idx start, regoff_t range, | ||
| 47 | struct re_registers *regs, | ||
| 48 | Idx stop, bool ret_len) internal_function; | ||
| 49 | static regoff_t re_search_stub (struct re_pattern_buffer *bufp, | ||
| 50 | const char *string, Idx length, Idx start, | ||
| 51 | regoff_t range, Idx stop, | ||
| 52 | struct re_registers *regs, | ||
| 53 | bool ret_len) internal_function; | ||
| 54 | static unsigned int re_copy_regs (struct re_registers *regs, regmatch_t *pmatch, | ||
| 55 | Idx nregs, int regs_allocated) | ||
| 56 | internal_function; | ||
| 57 | static reg_errcode_t prune_impossible_nodes (re_match_context_t *mctx) | ||
| 58 | internal_function; | ||
| 59 | static Idx check_matching (re_match_context_t *mctx, bool fl_longest_match, | ||
| 60 | Idx *p_match_first) internal_function; | ||
| 61 | static Idx check_halt_state_context (const re_match_context_t *mctx, | ||
| 62 | const re_dfastate_t *state, Idx idx) | ||
| 63 | internal_function; | ||
| 64 | static void update_regs (const re_dfa_t *dfa, regmatch_t *pmatch, | ||
| 65 | regmatch_t *prev_idx_match, Idx cur_node, | ||
| 66 | Idx cur_idx, Idx nmatch) internal_function; | ||
| 67 | static reg_errcode_t push_fail_stack (struct re_fail_stack_t *fs, | ||
| 68 | Idx str_idx, Idx dest_node, Idx nregs, | ||
| 69 | regmatch_t *regs, | ||
| 70 | re_node_set *eps_via_nodes) | ||
| 71 | internal_function; | ||
| 72 | static reg_errcode_t set_regs (const regex_t *preg, | ||
| 73 | const re_match_context_t *mctx, | ||
| 74 | size_t nmatch, regmatch_t *pmatch, | ||
| 75 | bool fl_backtrack) internal_function; | ||
| 76 | static reg_errcode_t free_fail_stack_return (struct re_fail_stack_t *fs) | ||
| 77 | internal_function; | ||
| 78 | |||
| 79 | #ifdef RE_ENABLE_I18N | ||
| 80 | static int sift_states_iter_mb (const re_match_context_t *mctx, | ||
| 81 | re_sift_context_t *sctx, | ||
| 82 | Idx node_idx, Idx str_idx, Idx max_str_idx) | ||
| 83 | internal_function; | ||
| 84 | #endif /* RE_ENABLE_I18N */ | ||
| 85 | static reg_errcode_t sift_states_backward (const re_match_context_t *mctx, | ||
| 86 | re_sift_context_t *sctx) | ||
| 87 | internal_function; | ||
| 88 | static reg_errcode_t build_sifted_states (const re_match_context_t *mctx, | ||
| 89 | re_sift_context_t *sctx, Idx str_idx, | ||
| 90 | re_node_set *cur_dest) | ||
| 91 | internal_function; | ||
| 92 | static reg_errcode_t update_cur_sifted_state (const re_match_context_t *mctx, | ||
| 93 | re_sift_context_t *sctx, | ||
| 94 | Idx str_idx, | ||
| 95 | re_node_set *dest_nodes) | ||
| 96 | internal_function; | ||
| 97 | static reg_errcode_t add_epsilon_src_nodes (const re_dfa_t *dfa, | ||
| 98 | re_node_set *dest_nodes, | ||
| 99 | const re_node_set *candidates) | ||
| 100 | internal_function; | ||
| 101 | static bool check_dst_limits (const re_match_context_t *mctx, | ||
| 102 | const re_node_set *limits, | ||
| 103 | Idx dst_node, Idx dst_idx, Idx src_node, | ||
| 104 | Idx src_idx) internal_function; | ||
| 105 | static int check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, | ||
| 106 | int boundaries, Idx subexp_idx, | ||
| 107 | Idx from_node, Idx bkref_idx) | ||
| 108 | internal_function; | ||
| 109 | static int check_dst_limits_calc_pos (const re_match_context_t *mctx, | ||
| 110 | Idx limit, Idx subexp_idx, | ||
| 111 | Idx node, Idx str_idx, | ||
| 112 | Idx bkref_idx) internal_function; | ||
| 113 | static reg_errcode_t check_subexp_limits (const re_dfa_t *dfa, | ||
| 114 | re_node_set *dest_nodes, | ||
| 115 | const re_node_set *candidates, | ||
| 116 | re_node_set *limits, | ||
| 117 | struct re_backref_cache_entry *bkref_ents, | ||
| 118 | Idx str_idx) internal_function; | ||
| 119 | static reg_errcode_t sift_states_bkref (const re_match_context_t *mctx, | ||
| 120 | re_sift_context_t *sctx, | ||
| 121 | Idx str_idx, const re_node_set *candidates) | ||
| 122 | internal_function; | ||
| 123 | static reg_errcode_t merge_state_array (const re_dfa_t *dfa, | ||
| 124 | re_dfastate_t **dst, | ||
| 125 | re_dfastate_t **src, Idx num) | ||
| 126 | internal_function; | ||
| 127 | static re_dfastate_t *find_recover_state (reg_errcode_t *err, | ||
| 128 | re_match_context_t *mctx) internal_function; | ||
| 129 | static re_dfastate_t *transit_state (reg_errcode_t *err, | ||
| 130 | re_match_context_t *mctx, | ||
| 131 | re_dfastate_t *state) internal_function; | ||
| 132 | static re_dfastate_t *merge_state_with_log (reg_errcode_t *err, | ||
| 133 | re_match_context_t *mctx, | ||
| 134 | re_dfastate_t *next_state) | ||
| 135 | internal_function; | ||
| 136 | static reg_errcode_t check_subexp_matching_top (re_match_context_t *mctx, | ||
| 137 | re_node_set *cur_nodes, | ||
| 138 | Idx str_idx) internal_function; | ||
| 139 | #if 0 | ||
| 140 | static re_dfastate_t *transit_state_sb (reg_errcode_t *err, | ||
| 141 | re_match_context_t *mctx, | ||
| 142 | re_dfastate_t *pstate) | ||
| 143 | internal_function; | ||
| 144 | #endif | ||
| 145 | #ifdef RE_ENABLE_I18N | ||
| 146 | static reg_errcode_t transit_state_mb (re_match_context_t *mctx, | ||
| 147 | re_dfastate_t *pstate) | ||
| 148 | internal_function; | ||
| 149 | #endif /* RE_ENABLE_I18N */ | ||
| 150 | static reg_errcode_t transit_state_bkref (re_match_context_t *mctx, | ||
| 151 | const re_node_set *nodes) | ||
| 152 | internal_function; | ||
| 153 | static reg_errcode_t get_subexp (re_match_context_t *mctx, | ||
| 154 | Idx bkref_node, Idx bkref_str_idx) | ||
| 155 | internal_function; | ||
| 156 | static reg_errcode_t get_subexp_sub (re_match_context_t *mctx, | ||
| 157 | const re_sub_match_top_t *sub_top, | ||
| 158 | re_sub_match_last_t *sub_last, | ||
| 159 | Idx bkref_node, Idx bkref_str) | ||
| 160 | internal_function; | ||
| 161 | static Idx find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes, | ||
| 162 | Idx subexp_idx, int type) internal_function; | ||
| 163 | static reg_errcode_t check_arrival (re_match_context_t *mctx, | ||
| 164 | state_array_t *path, Idx top_node, | ||
| 165 | Idx top_str, Idx last_node, Idx last_str, | ||
| 166 | int type) internal_function; | ||
| 167 | static reg_errcode_t check_arrival_add_next_nodes (re_match_context_t *mctx, | ||
| 168 | Idx str_idx, | ||
| 169 | re_node_set *cur_nodes, | ||
| 170 | re_node_set *next_nodes) | ||
| 171 | internal_function; | ||
| 172 | static reg_errcode_t check_arrival_expand_ecl (const re_dfa_t *dfa, | ||
| 173 | re_node_set *cur_nodes, | ||
| 174 | Idx ex_subexp, int type) | ||
| 175 | internal_function; | ||
| 176 | static reg_errcode_t check_arrival_expand_ecl_sub (const re_dfa_t *dfa, | ||
| 177 | re_node_set *dst_nodes, | ||
| 178 | Idx target, Idx ex_subexp, | ||
| 179 | int type) internal_function; | ||
| 180 | static reg_errcode_t expand_bkref_cache (re_match_context_t *mctx, | ||
| 181 | re_node_set *cur_nodes, Idx cur_str, | ||
| 182 | Idx subexp_num, int type) | ||
| 183 | internal_function; | ||
| 184 | static bool build_trtable (const re_dfa_t *dfa, | ||
| 185 | re_dfastate_t *state) internal_function; | ||
| 186 | #ifdef RE_ENABLE_I18N | ||
| 187 | static int check_node_accept_bytes (const re_dfa_t *dfa, Idx node_idx, | ||
| 188 | const re_string_t *input, Idx idx) | ||
| 189 | internal_function; | ||
| 190 | # ifdef _LIBC | ||
| 191 | static unsigned int find_collation_sequence_value (const unsigned char *mbs, | ||
| 192 | size_t name_len) | ||
| 193 | internal_function; | ||
| 194 | # endif /* _LIBC */ | ||
| 195 | #endif /* RE_ENABLE_I18N */ | ||
| 196 | static Idx group_nodes_into_DFAstates (const re_dfa_t *dfa, | ||
| 197 | const re_dfastate_t *state, | ||
| 198 | re_node_set *states_node, | ||
| 199 | bitset_t *states_ch) internal_function; | ||
| 200 | static bool check_node_accept (const re_match_context_t *mctx, | ||
| 201 | const re_token_t *node, Idx idx) | ||
| 202 | internal_function; | ||
| 203 | static reg_errcode_t extend_buffers (re_match_context_t *mctx) | ||
| 204 | internal_function; | ||
| 205 | |||
| 206 | /* Entry point for POSIX code. */ | ||
| 207 | |||
| 208 | /* regexec searches for a given pattern, specified by PREG, in the | ||
| 209 | string STRING. | ||
| 210 | |||
| 211 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | ||
| 212 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | ||
| 213 | least NMATCH elements, and we set them to the offsets of the | ||
| 214 | corresponding matched substrings. | ||
| 215 | |||
| 216 | EFLAGS specifies `execution flags' which affect matching: if | ||
| 217 | REG_NOTBOL is set, then ^ does not match at the beginning of the | ||
| 218 | string; if REG_NOTEOL is set, then $ does not match at the end. | ||
| 219 | |||
| 220 | We return 0 if we find a match and REG_NOMATCH if not. */ | ||
| 221 | |||
| 222 | int | ||
| 223 | regexec (preg, string, nmatch, pmatch, eflags) | ||
| 224 | const regex_t *__restrict preg; | ||
| 225 | const char *__restrict string; | ||
| 226 | size_t nmatch; | ||
| 227 | regmatch_t pmatch[]; | ||
| 228 | int eflags; | ||
| 229 | { | ||
| 230 | reg_errcode_t err; | ||
| 231 | Idx start, length; | ||
| 232 | #ifdef _LIBC | ||
| 233 | re_dfa_t *dfa = (re_dfa_t *) preg->buffer; | ||
| 234 | #endif | ||
| 235 | |||
| 236 | if (eflags & ~(REG_NOTBOL | REG_NOTEOL | REG_STARTEND)) | ||
| 237 | return REG_BADPAT; | ||
| 238 | |||
| 239 | if (eflags & REG_STARTEND) | ||
| 240 | { | ||
| 241 | start = pmatch[0].rm_so; | ||
| 242 | length = pmatch[0].rm_eo; | ||
| 243 | } | ||
| 244 | else | ||
| 245 | { | ||
| 246 | start = 0; | ||
| 247 | length = strlen (string); | ||
| 248 | } | ||
| 249 | |||
| 250 | __libc_lock_lock (dfa->lock); | ||
| 251 | if (preg->no_sub) | ||
| 252 | err = re_search_internal (preg, string, length, start, length, | ||
| 253 | length, 0, NULL, eflags); | ||
| 254 | else | ||
| 255 | err = re_search_internal (preg, string, length, start, length, | ||
| 256 | length, nmatch, pmatch, eflags); | ||
| 257 | __libc_lock_unlock (dfa->lock); | ||
| 258 | return err != REG_NOERROR; | ||
| 259 | } | ||
| 260 | |||
| 261 | #ifdef _LIBC | ||
| 262 | # include <shlib-compat.h> | ||
| 263 | versioned_symbol (libc, __regexec, regexec, GLIBC_2_3_4); | ||
| 264 | |||
| 265 | # if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4) | ||
| 266 | __typeof__ (__regexec) __compat_regexec; | ||
| 267 | |||
| 268 | int | ||
| 269 | attribute_compat_text_section | ||
| 270 | __compat_regexec (const regex_t *__restrict preg, | ||
| 271 | const char *__restrict string, size_t nmatch, | ||
| 272 | regmatch_t pmatch[], int eflags) | ||
| 273 | { | ||
| 274 | return regexec (preg, string, nmatch, pmatch, | ||
| 275 | eflags & (REG_NOTBOL | REG_NOTEOL)); | ||
| 276 | } | ||
| 277 | compat_symbol (libc, __compat_regexec, regexec, GLIBC_2_0); | ||
| 278 | # endif | ||
| 279 | #endif | ||
| 280 | |||
| 281 | /* Entry points for GNU code. */ | ||
| 282 | |||
| 283 | /* re_match, re_search, re_match_2, re_search_2 | ||
| 284 | |||
| 285 | The former two functions operate on STRING with length LENGTH, | ||
| 286 | while the later two operate on concatenation of STRING1 and STRING2 | ||
| 287 | with lengths LENGTH1 and LENGTH2, respectively. | ||
| 288 | |||
| 289 | re_match() matches the compiled pattern in BUFP against the string, | ||
| 290 | starting at index START. | ||
| 291 | |||
| 292 | re_search() first tries matching at index START, then it tries to match | ||
| 293 | starting from index START + 1, and so on. The last start position tried | ||
| 294 | is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same | ||
| 295 | way as re_match().) | ||
| 296 | |||
| 297 | The parameter STOP of re_{match,search}_2 specifies that no match exceeding | ||
| 298 | the first STOP characters of the concatenation of the strings should be | ||
| 299 | concerned. | ||
| 300 | |||
| 301 | If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match | ||
| 302 | and all groups is stored in REGS. (For the "_2" variants, the offsets are | ||
| 303 | computed relative to the concatenation, not relative to the individual | ||
| 304 | strings.) | ||
| 305 | |||
| 306 | On success, re_match* functions return the length of the match, re_search* | ||
| 307 | return the position of the start of the match. Return value -1 means no | ||
| 308 | match was found and -2 indicates an internal error. */ | ||
| 309 | |||
| 310 | regoff_t | ||
| 311 | re_match (bufp, string, length, start, regs) | ||
| 312 | struct re_pattern_buffer *bufp; | ||
| 313 | const char *string; | ||
| 314 | Idx length, start; | ||
| 315 | struct re_registers *regs; | ||
| 316 | { | ||
| 317 | return re_search_stub (bufp, string, length, start, 0, length, regs, true); | ||
| 318 | } | ||
| 319 | #ifdef _LIBC | ||
| 320 | weak_alias (__re_match, re_match) | ||
| 321 | #endif | ||
| 322 | |||
| 323 | regoff_t | ||
| 324 | re_search (bufp, string, length, start, range, regs) | ||
| 325 | struct re_pattern_buffer *bufp; | ||
| 326 | const char *string; | ||
| 327 | Idx length, start; | ||
| 328 | regoff_t range; | ||
| 329 | struct re_registers *regs; | ||
| 330 | { | ||
| 331 | return re_search_stub (bufp, string, length, start, range, length, regs, | ||
| 332 | false); | ||
| 333 | } | ||
| 334 | #ifdef _LIBC | ||
| 335 | weak_alias (__re_search, re_search) | ||
| 336 | #endif | ||
| 337 | |||
| 338 | regoff_t | ||
| 339 | re_match_2 (bufp, string1, length1, string2, length2, start, regs, stop) | ||
| 340 | struct re_pattern_buffer *bufp; | ||
| 341 | const char *string1, *string2; | ||
| 342 | Idx length1, length2, start, stop; | ||
| 343 | struct re_registers *regs; | ||
| 344 | { | ||
| 345 | return re_search_2_stub (bufp, string1, length1, string2, length2, | ||
| 346 | start, 0, regs, stop, true); | ||
| 347 | } | ||
| 348 | #ifdef _LIBC | ||
| 349 | weak_alias (__re_match_2, re_match_2) | ||
| 350 | #endif | ||
| 351 | |||
| 352 | regoff_t | ||
| 353 | re_search_2 (bufp, string1, length1, string2, length2, start, range, regs, stop) | ||
| 354 | struct re_pattern_buffer *bufp; | ||
| 355 | const char *string1, *string2; | ||
| 356 | Idx length1, length2, start, stop; | ||
| 357 | regoff_t range; | ||
| 358 | struct re_registers *regs; | ||
| 359 | { | ||
| 360 | return re_search_2_stub (bufp, string1, length1, string2, length2, | ||
| 361 | start, range, regs, stop, false); | ||
| 362 | } | ||
| 363 | #ifdef _LIBC | ||
| 364 | weak_alias (__re_search_2, re_search_2) | ||
| 365 | #endif | ||
| 366 | |||
| 367 | static regoff_t | ||
| 368 | internal_function | ||
| 369 | re_search_2_stub (struct re_pattern_buffer *bufp, | ||
| 370 | const char *string1, Idx length1, | ||
| 371 | const char *string2, Idx length2, | ||
| 372 | Idx start, regoff_t range, struct re_registers *regs, | ||
| 373 | Idx stop, bool ret_len) | ||
| 374 | { | ||
| 375 | const char *str; | ||
| 376 | regoff_t rval; | ||
| 377 | Idx len = length1 + length2; | ||
| 378 | char *s = NULL; | ||
| 379 | |||
| 380 | if (BE (length1 < 0 || length2 < 0 || stop < 0 || len < length1, 0)) | ||
| 381 | return -2; | ||
| 382 | |||
| 383 | /* Concatenate the strings. */ | ||
| 384 | if (length2 > 0) | ||
| 385 | if (length1 > 0) | ||
| 386 | { | ||
| 387 | s = re_malloc (char, len); | ||
| 388 | |||
| 389 | if (BE (s == NULL, 0)) | ||
| 390 | return -2; | ||
| 391 | #ifdef _LIBC | ||
| 392 | memcpy (__mempcpy (s, string1, length1), string2, length2); | ||
| 393 | #else | ||
| 394 | memcpy (s, string1, length1); | ||
| 395 | memcpy (s + length1, string2, length2); | ||
| 396 | #endif | ||
| 397 | str = s; | ||
| 398 | } | ||
| 399 | else | ||
| 400 | str = string2; | ||
| 401 | else | ||
| 402 | str = string1; | ||
| 403 | |||
| 404 | rval = re_search_stub (bufp, str, len, start, range, stop, regs, | ||
| 405 | ret_len); | ||
| 406 | re_free (s); | ||
| 407 | return rval; | ||
| 408 | } | ||
| 409 | |||
| 410 | /* The parameters have the same meaning as those of re_search. | ||
| 411 | Additional parameters: | ||
| 412 | If RET_LEN is true the length of the match is returned (re_match style); | ||
| 413 | otherwise the position of the match is returned. */ | ||
| 414 | |||
| 415 | static regoff_t | ||
| 416 | internal_function | ||
| 417 | re_search_stub (struct re_pattern_buffer *bufp, | ||
| 418 | const char *string, Idx length, | ||
| 419 | Idx start, regoff_t range, Idx stop, struct re_registers *regs, | ||
| 420 | bool ret_len) | ||
| 421 | { | ||
| 422 | reg_errcode_t result; | ||
| 423 | regmatch_t *pmatch; | ||
| 424 | Idx nregs; | ||
| 425 | regoff_t rval; | ||
| 426 | int eflags = 0; | ||
| 427 | #ifdef _LIBC | ||
| 428 | re_dfa_t *dfa = (re_dfa_t *) bufp->buffer; | ||
| 429 | #endif | ||
| 430 | Idx last_start = start + range; | ||
| 431 | |||
| 432 | /* Check for out-of-range. */ | ||
| 433 | if (BE (start < 0 || start > length, 0)) | ||
| 434 | return -1; | ||
| 435 | if (BE (length < last_start || (0 <= range && last_start < start), 0)) | ||
| 436 | last_start = length; | ||
| 437 | else if (BE (last_start < 0 || (range < 0 && start <= last_start), 0)) | ||
| 438 | last_start = 0; | ||
| 439 | |||
| 440 | __libc_lock_lock (dfa->lock); | ||
| 441 | |||
| 442 | eflags |= (bufp->not_bol) ? REG_NOTBOL : 0; | ||
| 443 | eflags |= (bufp->not_eol) ? REG_NOTEOL : 0; | ||
| 444 | |||
| 445 | /* Compile fastmap if we haven't yet. */ | ||
| 446 | if (start < last_start && bufp->fastmap != NULL && !bufp->fastmap_accurate) | ||
| 447 | re_compile_fastmap (bufp); | ||
| 448 | |||
| 449 | if (BE (bufp->no_sub, 0)) | ||
| 450 | regs = NULL; | ||
| 451 | |||
| 452 | /* We need at least 1 register. */ | ||
| 453 | if (regs == NULL) | ||
| 454 | nregs = 1; | ||
| 455 | else if (BE (bufp->regs_allocated == REGS_FIXED | ||
| 456 | && regs->num_regs <= bufp->re_nsub, 0)) | ||
| 457 | { | ||
| 458 | nregs = regs->num_regs; | ||
| 459 | if (BE (nregs < 1, 0)) | ||
| 460 | { | ||
| 461 | /* Nothing can be copied to regs. */ | ||
| 462 | regs = NULL; | ||
| 463 | nregs = 1; | ||
| 464 | } | ||
| 465 | } | ||
| 466 | else | ||
| 467 | nregs = bufp->re_nsub + 1; | ||
| 468 | pmatch = re_malloc (regmatch_t, nregs); | ||
| 469 | if (BE (pmatch == NULL, 0)) | ||
| 470 | { | ||
| 471 | rval = -2; | ||
| 472 | goto out; | ||
| 473 | } | ||
| 474 | |||
| 475 | result = re_search_internal (bufp, string, length, start, last_start, stop, | ||
| 476 | nregs, pmatch, eflags); | ||
| 477 | |||
| 478 | rval = 0; | ||
| 479 | |||
| 480 | /* I hope we needn't fill ther regs with -1's when no match was found. */ | ||
| 481 | if (result != REG_NOERROR) | ||
| 482 | rval = -1; | ||
| 483 | else if (regs != NULL) | ||
| 484 | { | ||
| 485 | /* If caller wants register contents data back, copy them. */ | ||
| 486 | bufp->regs_allocated = re_copy_regs (regs, pmatch, nregs, | ||
| 487 | bufp->regs_allocated); | ||
| 488 | if (BE (bufp->regs_allocated == REGS_UNALLOCATED, 0)) | ||
| 489 | rval = -2; | ||
| 490 | } | ||
| 491 | |||
| 492 | if (BE (rval == 0, 1)) | ||
| 493 | { | ||
| 494 | if (ret_len) | ||
| 495 | { | ||
| 496 | assert (pmatch[0].rm_so == start); | ||
| 497 | rval = pmatch[0].rm_eo - start; | ||
| 498 | } | ||
| 499 | else | ||
| 500 | rval = pmatch[0].rm_so; | ||
| 501 | } | ||
| 502 | re_free (pmatch); | ||
| 503 | out: | ||
| 504 | __libc_lock_unlock (dfa->lock); | ||
| 505 | return rval; | ||
| 506 | } | ||
| 507 | |||
| 508 | static unsigned int | ||
| 509 | internal_function | ||
| 510 | re_copy_regs (struct re_registers *regs, regmatch_t *pmatch, Idx nregs, | ||
| 511 | int regs_allocated) | ||
| 512 | { | ||
| 513 | int rval = REGS_REALLOCATE; | ||
| 514 | Idx i; | ||
| 515 | Idx need_regs = nregs + 1; | ||
| 516 | /* We need one extra element beyond `num_regs' for the `-1' marker GNU code | ||
| 517 | uses. */ | ||
| 518 | |||
| 519 | /* Have the register data arrays been allocated? */ | ||
| 520 | if (regs_allocated == REGS_UNALLOCATED) | ||
| 521 | { /* No. So allocate them with malloc. */ | ||
| 522 | regs->start = re_malloc (regoff_t, need_regs); | ||
| 523 | if (BE (regs->start == NULL, 0)) | ||
| 524 | return REGS_UNALLOCATED; | ||
| 525 | regs->end = re_malloc (regoff_t, need_regs); | ||
| 526 | if (BE (regs->end == NULL, 0)) | ||
| 527 | { | ||
| 528 | re_free (regs->start); | ||
| 529 | return REGS_UNALLOCATED; | ||
| 530 | } | ||
| 531 | regs->num_regs = need_regs; | ||
| 532 | } | ||
| 533 | else if (regs_allocated == REGS_REALLOCATE) | ||
| 534 | { /* Yes. If we need more elements than were already | ||
| 535 | allocated, reallocate them. If we need fewer, just | ||
| 536 | leave it alone. */ | ||
| 537 | if (BE (need_regs > regs->num_regs, 0)) | ||
| 538 | { | ||
| 539 | regoff_t *new_start = re_realloc (regs->start, regoff_t, need_regs); | ||
| 540 | regoff_t *new_end; | ||
| 541 | if (BE (new_start == NULL, 0)) | ||
| 542 | return REGS_UNALLOCATED; | ||
| 543 | new_end = re_realloc (regs->end, regoff_t, need_regs); | ||
| 544 | if (BE (new_end == NULL, 0)) | ||
| 545 | { | ||
| 546 | re_free (new_start); | ||
| 547 | return REGS_UNALLOCATED; | ||
| 548 | } | ||
| 549 | regs->start = new_start; | ||
| 550 | regs->end = new_end; | ||
| 551 | regs->num_regs = need_regs; | ||
| 552 | } | ||
| 553 | } | ||
| 554 | else | ||
| 555 | { | ||
| 556 | assert (regs_allocated == REGS_FIXED); | ||
| 557 | /* This function may not be called with REGS_FIXED and nregs too big. */ | ||
| 558 | assert (regs->num_regs >= nregs); | ||
| 559 | rval = REGS_FIXED; | ||
| 560 | } | ||
| 561 | |||
| 562 | /* Copy the regs. */ | ||
| 563 | for (i = 0; i < nregs; ++i) | ||
| 564 | { | ||
| 565 | regs->start[i] = pmatch[i].rm_so; | ||
| 566 | regs->end[i] = pmatch[i].rm_eo; | ||
| 567 | } | ||
| 568 | for ( ; i < regs->num_regs; ++i) | ||
| 569 | regs->start[i] = regs->end[i] = -1; | ||
| 570 | |||
| 571 | return rval; | ||
| 572 | } | ||
| 573 | |||
| 574 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | ||
| 575 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | ||
| 576 | this memory for recording register information. STARTS and ENDS | ||
| 577 | must be allocated using the malloc library routine, and must each | ||
| 578 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | ||
| 579 | |||
| 580 | If NUM_REGS == 0, then subsequent matches should allocate their own | ||
| 581 | register data. | ||
| 582 | |||
| 583 | Unless this function is called, the first search or match using | ||
| 584 | PATTERN_BUFFER will allocate its own register data, without | ||
| 585 | freeing the old data. */ | ||
| 586 | |||
| 587 | void | ||
| 588 | re_set_registers (bufp, regs, num_regs, starts, ends) | ||
| 589 | struct re_pattern_buffer *bufp; | ||
| 590 | struct re_registers *regs; | ||
| 591 | __re_size_t num_regs; | ||
| 592 | regoff_t *starts, *ends; | ||
| 593 | { | ||
| 594 | if (num_regs) | ||
| 595 | { | ||
| 596 | bufp->regs_allocated = REGS_REALLOCATE; | ||
| 597 | regs->num_regs = num_regs; | ||
| 598 | regs->start = starts; | ||
| 599 | regs->end = ends; | ||
| 600 | } | ||
| 601 | else | ||
| 602 | { | ||
| 603 | bufp->regs_allocated = REGS_UNALLOCATED; | ||
| 604 | regs->num_regs = 0; | ||
| 605 | regs->start = regs->end = NULL; | ||
| 606 | } | ||
| 607 | } | ||
| 608 | #ifdef _LIBC | ||
| 609 | weak_alias (__re_set_registers, re_set_registers) | ||
| 610 | #endif | ||
| 611 | |||
| 612 | /* Entry points compatible with 4.2 BSD regex library. We don't define | ||
| 613 | them unless specifically requested. */ | ||
| 614 | |||
| 615 | #if defined _REGEX_RE_COMP || defined _LIBC | ||
| 616 | int | ||
| 617 | # ifdef _LIBC | ||
| 618 | weak_function | ||
| 619 | # endif | ||
| 620 | re_exec (s) | ||
| 621 | const char *s; | ||
| 622 | { | ||
| 623 | return 0 == regexec (&re_comp_buf, s, 0, NULL, 0); | ||
| 624 | } | ||
| 625 | #endif /* _REGEX_RE_COMP */ | ||
| 626 | |||
| 627 | /* Internal entry point. */ | ||
| 628 | |||
| 629 | /* Searches for a compiled pattern PREG in the string STRING, whose | ||
| 630 | length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same | ||
| 631 | meaning as with regexec. LAST_START is START + RANGE, where | ||
| 632 | START and RANGE have the same meaning as with re_search. | ||
| 633 | Return REG_NOERROR if we find a match, and REG_NOMATCH if not, | ||
| 634 | otherwise return the error code. | ||
| 635 | Note: We assume front end functions already check ranges. | ||
| 636 | (0 <= LAST_START && LAST_START <= LENGTH) */ | ||
| 637 | |||
| 638 | static reg_errcode_t | ||
| 639 | internal_function | ||
| 640 | re_search_internal (const regex_t *preg, | ||
| 641 | const char *string, Idx length, | ||
| 642 | Idx start, Idx last_start, Idx stop, | ||
| 643 | size_t nmatch, regmatch_t pmatch[], | ||
| 644 | int eflags) | ||
| 645 | { | ||
| 646 | reg_errcode_t err; | ||
| 647 | const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer; | ||
| 648 | Idx left_lim, right_lim; | ||
| 649 | int incr; | ||
| 650 | bool fl_longest_match; | ||
| 651 | int match_kind; | ||
| 652 | Idx match_first; | ||
| 653 | Idx match_last = REG_MISSING; | ||
| 654 | Idx extra_nmatch; | ||
| 655 | bool sb; | ||
| 656 | int ch; | ||
| 657 | #if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L) | ||
| 658 | re_match_context_t mctx = { .dfa = dfa }; | ||
| 659 | #else | ||
| 660 | re_match_context_t mctx; | ||
| 661 | #endif | ||
| 662 | char *fastmap = ((preg->fastmap != NULL && preg->fastmap_accurate | ||
| 663 | && start != last_start && !preg->can_be_null) | ||
| 664 | ? preg->fastmap : NULL); | ||
| 665 | RE_TRANSLATE_TYPE t = preg->translate; | ||
| 666 | |||
| 667 | #if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)) | ||
| 668 | memset (&mctx, '\0', sizeof (re_match_context_t)); | ||
| 669 | mctx.dfa = dfa; | ||
| 670 | #endif | ||
| 671 | |||
| 672 | extra_nmatch = (nmatch > preg->re_nsub) ? nmatch - (preg->re_nsub + 1) : 0; | ||
| 673 | nmatch -= extra_nmatch; | ||
| 674 | |||
| 675 | /* Check if the DFA haven't been compiled. */ | ||
| 676 | if (BE (preg->used == 0 || dfa->init_state == NULL | ||
| 677 | || dfa->init_state_word == NULL || dfa->init_state_nl == NULL | ||
| 678 | || dfa->init_state_begbuf == NULL, 0)) | ||
| 679 | return REG_NOMATCH; | ||
| 680 | |||
| 681 | #ifdef DEBUG | ||
| 682 | /* We assume front-end functions already check them. */ | ||
| 683 | assert (0 <= last_start && last_start <= length); | ||
| 684 | #endif | ||
| 685 | |||
| 686 | /* If initial states with non-begbuf contexts have no elements, | ||
| 687 | the regex must be anchored. If preg->newline_anchor is set, | ||
| 688 | we'll never use init_state_nl, so do not check it. */ | ||
| 689 | if (dfa->init_state->nodes.nelem == 0 | ||
| 690 | && dfa->init_state_word->nodes.nelem == 0 | ||
| 691 | && (dfa->init_state_nl->nodes.nelem == 0 | ||
| 692 | || !preg->newline_anchor)) | ||
| 693 | { | ||
| 694 | if (start != 0 && last_start != 0) | ||
| 695 | return REG_NOMATCH; | ||
| 696 | start = last_start = 0; | ||
| 697 | } | ||
| 698 | |||
| 699 | /* We must check the longest matching, if nmatch > 0. */ | ||
| 700 | fl_longest_match = (nmatch != 0 || dfa->nbackref); | ||
| 701 | |||
| 702 | err = re_string_allocate (&mctx.input, string, length, dfa->nodes_len + 1, | ||
| 703 | preg->translate, preg->syntax & RE_ICASE, dfa); | ||
| 704 | if (BE (err != REG_NOERROR, 0)) | ||
| 705 | goto free_return; | ||
| 706 | mctx.input.stop = stop; | ||
| 707 | mctx.input.raw_stop = stop; | ||
| 708 | mctx.input.newline_anchor = preg->newline_anchor; | ||
| 709 | |||
| 710 | err = match_ctx_init (&mctx, eflags, dfa->nbackref * 2); | ||
| 711 | if (BE (err != REG_NOERROR, 0)) | ||
| 712 | goto free_return; | ||
| 713 | |||
| 714 | /* We will log all the DFA states through which the dfa pass, | ||
| 715 | if nmatch > 1, or this dfa has "multibyte node", which is a | ||
| 716 | back-reference or a node which can accept multibyte character or | ||
| 717 | multi character collating element. */ | ||
| 718 | if (nmatch > 1 || dfa->has_mb_node) | ||
| 719 | { | ||
| 720 | /* Avoid overflow. */ | ||
| 721 | if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= mctx.input.bufs_len, 0)) | ||
| 722 | { | ||
| 723 | err = REG_ESPACE; | ||
| 724 | goto free_return; | ||
| 725 | } | ||
| 726 | |||
| 727 | mctx.state_log = re_malloc (re_dfastate_t *, mctx.input.bufs_len + 1); | ||
| 728 | if (BE (mctx.state_log == NULL, 0)) | ||
| 729 | { | ||
| 730 | err = REG_ESPACE; | ||
| 731 | goto free_return; | ||
| 732 | } | ||
| 733 | } | ||
| 734 | else | ||
| 735 | mctx.state_log = NULL; | ||
| 736 | |||
| 737 | match_first = start; | ||
| 738 | mctx.input.tip_context = (eflags & REG_NOTBOL) ? CONTEXT_BEGBUF | ||
| 739 | : CONTEXT_NEWLINE | CONTEXT_BEGBUF; | ||
| 740 | |||
| 741 | /* Check incrementally whether of not the input string match. */ | ||
| 742 | incr = (last_start < start) ? -1 : 1; | ||
| 743 | left_lim = (last_start < start) ? last_start : start; | ||
| 744 | right_lim = (last_start < start) ? start : last_start; | ||
| 745 | sb = dfa->mb_cur_max == 1; | ||
| 746 | match_kind = | ||
| 747 | (fastmap | ||
| 748 | ? ((sb || !(preg->syntax & RE_ICASE || t) ? 4 : 0) | ||
| 749 | | (start <= last_start ? 2 : 0) | ||
| 750 | | (t != NULL ? 1 : 0)) | ||
| 751 | : 8); | ||
| 752 | |||
| 753 | for (;; match_first += incr) | ||
| 754 | { | ||
| 755 | err = REG_NOMATCH; | ||
| 756 | if (match_first < left_lim || right_lim < match_first) | ||
| 757 | goto free_return; | ||
| 758 | |||
| 759 | /* Advance as rapidly as possible through the string, until we | ||
| 760 | find a plausible place to start matching. This may be done | ||
| 761 | with varying efficiency, so there are various possibilities: | ||
| 762 | only the most common of them are specialized, in order to | ||
| 763 | save on code size. We use a switch statement for speed. */ | ||
| 764 | switch (match_kind) | ||
| 765 | { | ||
| 766 | case 8: | ||
| 767 | /* No fastmap. */ | ||
| 768 | break; | ||
| 769 | |||
| 770 | case 7: | ||
| 771 | /* Fastmap with single-byte translation, match forward. */ | ||
| 772 | while (BE (match_first < right_lim, 1) | ||
| 773 | && !fastmap[t[(unsigned char) string[match_first]]]) | ||
| 774 | ++match_first; | ||
| 775 | goto forward_match_found_start_or_reached_end; | ||
| 776 | |||
| 777 | case 6: | ||
| 778 | /* Fastmap without translation, match forward. */ | ||
| 779 | while (BE (match_first < right_lim, 1) | ||
| 780 | && !fastmap[(unsigned char) string[match_first]]) | ||
| 781 | ++match_first; | ||
| 782 | |||
| 783 | forward_match_found_start_or_reached_end: | ||
| 784 | if (BE (match_first == right_lim, 0)) | ||
| 785 | { | ||
| 786 | ch = match_first >= length | ||
| 787 | ? 0 : (unsigned char) string[match_first]; | ||
| 788 | if (!fastmap[t ? t[ch] : ch]) | ||
| 789 | goto free_return; | ||
| 790 | } | ||
| 791 | break; | ||
| 792 | |||
| 793 | case 4: | ||
| 794 | case 5: | ||
| 795 | /* Fastmap without multi-byte translation, match backwards. */ | ||
| 796 | while (match_first >= left_lim) | ||
| 797 | { | ||
| 798 | ch = match_first >= length | ||
| 799 | ? 0 : (unsigned char) string[match_first]; | ||
| 800 | if (fastmap[t ? t[ch] : ch]) | ||
| 801 | break; | ||
| 802 | --match_first; | ||
| 803 | } | ||
| 804 | if (match_first < left_lim) | ||
| 805 | goto free_return; | ||
| 806 | break; | ||
| 807 | |||
| 808 | default: | ||
| 809 | /* In this case, we can't determine easily the current byte, | ||
| 810 | since it might be a component byte of a multibyte | ||
| 811 | character. Then we use the constructed buffer instead. */ | ||
| 812 | for (;;) | ||
| 813 | { | ||
| 814 | /* If MATCH_FIRST is out of the valid range, reconstruct the | ||
| 815 | buffers. */ | ||
| 816 | __re_size_t offset = match_first - mctx.input.raw_mbs_idx; | ||
| 817 | if (BE (offset >= (__re_size_t) mctx.input.valid_raw_len, 0)) | ||
| 818 | { | ||
| 819 | err = re_string_reconstruct (&mctx.input, match_first, | ||
| 820 | eflags); | ||
| 821 | if (BE (err != REG_NOERROR, 0)) | ||
| 822 | goto free_return; | ||
| 823 | |||
| 824 | offset = match_first - mctx.input.raw_mbs_idx; | ||
| 825 | } | ||
| 826 | /* If MATCH_FIRST is out of the buffer, leave it as '\0'. | ||
| 827 | Note that MATCH_FIRST must not be smaller than 0. */ | ||
| 828 | ch = (match_first >= length | ||
| 829 | ? 0 : re_string_byte_at (&mctx.input, offset)); | ||
| 830 | if (fastmap[ch]) | ||
| 831 | break; | ||
| 832 | match_first += incr; | ||
| 833 | if (match_first < left_lim || match_first > right_lim) | ||
| 834 | { | ||
| 835 | err = REG_NOMATCH; | ||
| 836 | goto free_return; | ||
| 837 | } | ||
| 838 | } | ||
| 839 | break; | ||
| 840 | } | ||
| 841 | |||
| 842 | /* Reconstruct the buffers so that the matcher can assume that | ||
| 843 | the matching starts from the beginning of the buffer. */ | ||
| 844 | err = re_string_reconstruct (&mctx.input, match_first, eflags); | ||
| 845 | if (BE (err != REG_NOERROR, 0)) | ||
| 846 | goto free_return; | ||
| 847 | |||
| 848 | #ifdef RE_ENABLE_I18N | ||
| 849 | /* Don't consider this char as a possible match start if it part, | ||
| 850 | yet isn't the head, of a multibyte character. */ | ||
| 851 | if (!sb && !re_string_first_byte (&mctx.input, 0)) | ||
| 852 | continue; | ||
| 853 | #endif | ||
| 854 | |||
| 855 | /* It seems to be appropriate one, then use the matcher. */ | ||
| 856 | /* We assume that the matching starts from 0. */ | ||
| 857 | mctx.state_log_top = mctx.nbkref_ents = mctx.max_mb_elem_len = 0; | ||
| 858 | match_last = check_matching (&mctx, fl_longest_match, | ||
| 859 | start <= last_start ? &match_first : NULL); | ||
| 860 | if (match_last != REG_MISSING) | ||
| 861 | { | ||
| 862 | if (BE (match_last == REG_ERROR, 0)) | ||
| 863 | { | ||
| 864 | err = REG_ESPACE; | ||
| 865 | goto free_return; | ||
| 866 | } | ||
| 867 | else | ||
| 868 | { | ||
| 869 | mctx.match_last = match_last; | ||
| 870 | if ((!preg->no_sub && nmatch > 1) || dfa->nbackref) | ||
| 871 | { | ||
| 872 | re_dfastate_t *pstate = mctx.state_log[match_last]; | ||
| 873 | mctx.last_node = check_halt_state_context (&mctx, pstate, | ||
| 874 | match_last); | ||
| 875 | } | ||
| 876 | if ((!preg->no_sub && nmatch > 1 && dfa->has_plural_match) | ||
| 877 | || dfa->nbackref) | ||
| 878 | { | ||
| 879 | err = prune_impossible_nodes (&mctx); | ||
| 880 | if (err == REG_NOERROR) | ||
| 881 | break; | ||
| 882 | if (BE (err != REG_NOMATCH, 0)) | ||
| 883 | goto free_return; | ||
| 884 | match_last = REG_MISSING; | ||
| 885 | } | ||
| 886 | else | ||
| 887 | break; /* We found a match. */ | ||
| 888 | } | ||
| 889 | } | ||
| 890 | |||
| 891 | match_ctx_clean (&mctx); | ||
| 892 | } | ||
| 893 | |||
| 894 | #ifdef DEBUG | ||
| 895 | assert (match_last != REG_MISSING); | ||
| 896 | assert (err == REG_NOERROR); | ||
| 897 | #endif | ||
| 898 | |||
| 899 | /* Set pmatch[] if we need. */ | ||
| 900 | if (nmatch > 0) | ||
| 901 | { | ||
| 902 | Idx reg_idx; | ||
| 903 | |||
| 904 | /* Initialize registers. */ | ||
| 905 | for (reg_idx = 1; reg_idx < nmatch; ++reg_idx) | ||
| 906 | pmatch[reg_idx].rm_so = pmatch[reg_idx].rm_eo = -1; | ||
| 907 | |||
| 908 | /* Set the points where matching start/end. */ | ||
| 909 | pmatch[0].rm_so = 0; | ||
| 910 | pmatch[0].rm_eo = mctx.match_last; | ||
| 911 | /* FIXME: This function should fail if mctx.match_last exceeds | ||
| 912 | the maximum possible regoff_t value. We need a new error | ||
| 913 | code REG_OVERFLOW. */ | ||
| 914 | |||
| 915 | if (!preg->no_sub && nmatch > 1) | ||
| 916 | { | ||
| 917 | err = set_regs (preg, &mctx, nmatch, pmatch, | ||
| 918 | dfa->has_plural_match && dfa->nbackref > 0); | ||
| 919 | if (BE (err != REG_NOERROR, 0)) | ||
| 920 | goto free_return; | ||
| 921 | } | ||
| 922 | |||
| 923 | /* At last, add the offset to the each registers, since we slided | ||
| 924 | the buffers so that we could assume that the matching starts | ||
| 925 | from 0. */ | ||
| 926 | for (reg_idx = 0; reg_idx < nmatch; ++reg_idx) | ||
| 927 | if (pmatch[reg_idx].rm_so != -1) | ||
| 928 | { | ||
| 929 | #ifdef RE_ENABLE_I18N | ||
| 930 | if (BE (mctx.input.offsets_needed != 0, 0)) | ||
| 931 | { | ||
| 932 | pmatch[reg_idx].rm_so = | ||
| 933 | (pmatch[reg_idx].rm_so == mctx.input.valid_len | ||
| 934 | ? mctx.input.valid_raw_len | ||
| 935 | : mctx.input.offsets[pmatch[reg_idx].rm_so]); | ||
| 936 | pmatch[reg_idx].rm_eo = | ||
| 937 | (pmatch[reg_idx].rm_eo == mctx.input.valid_len | ||
| 938 | ? mctx.input.valid_raw_len | ||
| 939 | : mctx.input.offsets[pmatch[reg_idx].rm_eo]); | ||
| 940 | } | ||
| 941 | #else | ||
| 942 | assert (mctx.input.offsets_needed == 0); | ||
| 943 | #endif | ||
| 944 | pmatch[reg_idx].rm_so += match_first; | ||
| 945 | pmatch[reg_idx].rm_eo += match_first; | ||
| 946 | } | ||
| 947 | for (reg_idx = 0; reg_idx < extra_nmatch; ++reg_idx) | ||
| 948 | { | ||
| 949 | pmatch[nmatch + reg_idx].rm_so = -1; | ||
| 950 | pmatch[nmatch + reg_idx].rm_eo = -1; | ||
| 951 | } | ||
| 952 | |||
| 953 | if (dfa->subexp_map) | ||
| 954 | for (reg_idx = 0; reg_idx + 1 < nmatch; reg_idx++) | ||
| 955 | if (dfa->subexp_map[reg_idx] != reg_idx) | ||
| 956 | { | ||
| 957 | pmatch[reg_idx + 1].rm_so | ||
| 958 | = pmatch[dfa->subexp_map[reg_idx] + 1].rm_so; | ||
| 959 | pmatch[reg_idx + 1].rm_eo | ||
| 960 | = pmatch[dfa->subexp_map[reg_idx] + 1].rm_eo; | ||
| 961 | } | ||
| 962 | } | ||
| 963 | |||
| 964 | free_return: | ||
| 965 | re_free (mctx.state_log); | ||
| 966 | if (dfa->nbackref) | ||
| 967 | match_ctx_free (&mctx); | ||
| 968 | re_string_destruct (&mctx.input); | ||
| 969 | return err; | ||
| 970 | } | ||
| 971 | |||
| 972 | static reg_errcode_t | ||
| 973 | internal_function | ||
| 974 | prune_impossible_nodes (re_match_context_t *mctx) | ||
| 975 | { | ||
| 976 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 977 | Idx halt_node, match_last; | ||
| 978 | reg_errcode_t ret; | ||
| 979 | re_dfastate_t **sifted_states; | ||
| 980 | re_dfastate_t **lim_states = NULL; | ||
| 981 | re_sift_context_t sctx; | ||
| 982 | #ifdef DEBUG | ||
| 983 | assert (mctx->state_log != NULL); | ||
| 984 | #endif | ||
| 985 | match_last = mctx->match_last; | ||
| 986 | halt_node = mctx->last_node; | ||
| 987 | |||
| 988 | /* Avoid overflow. */ | ||
| 989 | if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= match_last, 0)) | ||
| 990 | return REG_ESPACE; | ||
| 991 | |||
| 992 | sifted_states = re_malloc (re_dfastate_t *, match_last + 1); | ||
| 993 | if (BE (sifted_states == NULL, 0)) | ||
| 994 | { | ||
| 995 | ret = REG_ESPACE; | ||
| 996 | goto free_return; | ||
| 997 | } | ||
| 998 | if (dfa->nbackref) | ||
| 999 | { | ||
| 1000 | lim_states = re_malloc (re_dfastate_t *, match_last + 1); | ||
| 1001 | if (BE (lim_states == NULL, 0)) | ||
| 1002 | { | ||
| 1003 | ret = REG_ESPACE; | ||
| 1004 | goto free_return; | ||
| 1005 | } | ||
| 1006 | while (1) | ||
| 1007 | { | ||
| 1008 | memset (lim_states, '\0', | ||
| 1009 | sizeof (re_dfastate_t *) * (match_last + 1)); | ||
| 1010 | sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, | ||
| 1011 | match_last); | ||
| 1012 | ret = sift_states_backward (mctx, &sctx); | ||
| 1013 | re_node_set_free (&sctx.limits); | ||
| 1014 | if (BE (ret != REG_NOERROR, 0)) | ||
| 1015 | goto free_return; | ||
| 1016 | if (sifted_states[0] != NULL || lim_states[0] != NULL) | ||
| 1017 | break; | ||
| 1018 | do | ||
| 1019 | { | ||
| 1020 | --match_last; | ||
| 1021 | if (! REG_VALID_INDEX (match_last)) | ||
| 1022 | { | ||
| 1023 | ret = REG_NOMATCH; | ||
| 1024 | goto free_return; | ||
| 1025 | } | ||
| 1026 | } while (mctx->state_log[match_last] == NULL | ||
| 1027 | || !mctx->state_log[match_last]->halt); | ||
| 1028 | halt_node = check_halt_state_context (mctx, | ||
| 1029 | mctx->state_log[match_last], | ||
| 1030 | match_last); | ||
| 1031 | } | ||
| 1032 | ret = merge_state_array (dfa, sifted_states, lim_states, | ||
| 1033 | match_last + 1); | ||
| 1034 | re_free (lim_states); | ||
| 1035 | lim_states = NULL; | ||
| 1036 | if (BE (ret != REG_NOERROR, 0)) | ||
| 1037 | goto free_return; | ||
| 1038 | } | ||
| 1039 | else | ||
| 1040 | { | ||
| 1041 | sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, match_last); | ||
| 1042 | ret = sift_states_backward (mctx, &sctx); | ||
| 1043 | re_node_set_free (&sctx.limits); | ||
| 1044 | if (BE (ret != REG_NOERROR, 0)) | ||
| 1045 | goto free_return; | ||
| 1046 | } | ||
| 1047 | re_free (mctx->state_log); | ||
| 1048 | mctx->state_log = sifted_states; | ||
| 1049 | sifted_states = NULL; | ||
| 1050 | mctx->last_node = halt_node; | ||
| 1051 | mctx->match_last = match_last; | ||
| 1052 | ret = REG_NOERROR; | ||
| 1053 | free_return: | ||
| 1054 | re_free (sifted_states); | ||
| 1055 | re_free (lim_states); | ||
| 1056 | return ret; | ||
| 1057 | } | ||
| 1058 | |||
| 1059 | /* Acquire an initial state and return it. | ||
| 1060 | We must select appropriate initial state depending on the context, | ||
| 1061 | since initial states may have constraints like "\<", "^", etc.. */ | ||
| 1062 | |||
| 1063 | static inline re_dfastate_t * | ||
| 1064 | __attribute ((always_inline)) internal_function | ||
| 1065 | acquire_init_state_context (reg_errcode_t *err, const re_match_context_t *mctx, | ||
| 1066 | Idx idx) | ||
| 1067 | { | ||
| 1068 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1069 | if (dfa->init_state->has_constraint) | ||
| 1070 | { | ||
| 1071 | unsigned int context; | ||
| 1072 | context = re_string_context_at (&mctx->input, idx - 1, mctx->eflags); | ||
| 1073 | if (IS_WORD_CONTEXT (context)) | ||
| 1074 | return dfa->init_state_word; | ||
| 1075 | else if (IS_ORDINARY_CONTEXT (context)) | ||
| 1076 | return dfa->init_state; | ||
| 1077 | else if (IS_BEGBUF_CONTEXT (context) && IS_NEWLINE_CONTEXT (context)) | ||
| 1078 | return dfa->init_state_begbuf; | ||
| 1079 | else if (IS_NEWLINE_CONTEXT (context)) | ||
| 1080 | return dfa->init_state_nl; | ||
| 1081 | else if (IS_BEGBUF_CONTEXT (context)) | ||
| 1082 | { | ||
| 1083 | /* It is relatively rare case, then calculate on demand. */ | ||
| 1084 | return re_acquire_state_context (err, dfa, | ||
| 1085 | dfa->init_state->entrance_nodes, | ||
| 1086 | context); | ||
| 1087 | } | ||
| 1088 | else | ||
| 1089 | /* Must not happen? */ | ||
| 1090 | return dfa->init_state; | ||
| 1091 | } | ||
| 1092 | else | ||
| 1093 | return dfa->init_state; | ||
| 1094 | } | ||
| 1095 | |||
| 1096 | /* Check whether the regular expression match input string INPUT or not, | ||
| 1097 | and return the index where the matching end. Return REG_MISSING if | ||
| 1098 | there is no match, and return REG_ERROR in case of an error. | ||
| 1099 | FL_LONGEST_MATCH means we want the POSIX longest matching. | ||
| 1100 | If P_MATCH_FIRST is not NULL, and the match fails, it is set to the | ||
| 1101 | next place where we may want to try matching. | ||
| 1102 | Note that the matcher assume that the maching starts from the current | ||
| 1103 | index of the buffer. */ | ||
| 1104 | |||
| 1105 | static Idx | ||
| 1106 | internal_function | ||
| 1107 | check_matching (re_match_context_t *mctx, bool fl_longest_match, | ||
| 1108 | Idx *p_match_first) | ||
| 1109 | { | ||
| 1110 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1111 | reg_errcode_t err; | ||
| 1112 | Idx match = 0; | ||
| 1113 | Idx match_last = REG_MISSING; | ||
| 1114 | Idx cur_str_idx = re_string_cur_idx (&mctx->input); | ||
| 1115 | re_dfastate_t *cur_state; | ||
| 1116 | bool at_init_state = p_match_first != NULL; | ||
| 1117 | Idx next_start_idx = cur_str_idx; | ||
| 1118 | |||
| 1119 | err = REG_NOERROR; | ||
| 1120 | cur_state = acquire_init_state_context (&err, mctx, cur_str_idx); | ||
| 1121 | /* An initial state must not be NULL (invalid). */ | ||
| 1122 | if (BE (cur_state == NULL, 0)) | ||
| 1123 | { | ||
| 1124 | assert (err == REG_ESPACE); | ||
| 1125 | return REG_ERROR; | ||
| 1126 | } | ||
| 1127 | |||
| 1128 | if (mctx->state_log != NULL) | ||
| 1129 | { | ||
| 1130 | mctx->state_log[cur_str_idx] = cur_state; | ||
| 1131 | |||
| 1132 | /* Check OP_OPEN_SUBEXP in the initial state in case that we use them | ||
| 1133 | later. E.g. Processing back references. */ | ||
| 1134 | if (BE (dfa->nbackref, 0)) | ||
| 1135 | { | ||
| 1136 | at_init_state = false; | ||
| 1137 | err = check_subexp_matching_top (mctx, &cur_state->nodes, 0); | ||
| 1138 | if (BE (err != REG_NOERROR, 0)) | ||
| 1139 | return err; | ||
| 1140 | |||
| 1141 | if (cur_state->has_backref) | ||
| 1142 | { | ||
| 1143 | err = transit_state_bkref (mctx, &cur_state->nodes); | ||
| 1144 | if (BE (err != REG_NOERROR, 0)) | ||
| 1145 | return err; | ||
| 1146 | } | ||
| 1147 | } | ||
| 1148 | } | ||
| 1149 | |||
| 1150 | /* If the RE accepts NULL string. */ | ||
| 1151 | if (BE (cur_state->halt, 0)) | ||
| 1152 | { | ||
| 1153 | if (!cur_state->has_constraint | ||
| 1154 | || check_halt_state_context (mctx, cur_state, cur_str_idx)) | ||
| 1155 | { | ||
| 1156 | if (!fl_longest_match) | ||
| 1157 | return cur_str_idx; | ||
| 1158 | else | ||
| 1159 | { | ||
| 1160 | match_last = cur_str_idx; | ||
| 1161 | match = 1; | ||
| 1162 | } | ||
| 1163 | } | ||
| 1164 | } | ||
| 1165 | |||
| 1166 | while (!re_string_eoi (&mctx->input)) | ||
| 1167 | { | ||
| 1168 | re_dfastate_t *old_state = cur_state; | ||
| 1169 | Idx next_char_idx = re_string_cur_idx (&mctx->input) + 1; | ||
| 1170 | |||
| 1171 | if (BE (next_char_idx >= mctx->input.bufs_len, 0) | ||
| 1172 | || (BE (next_char_idx >= mctx->input.valid_len, 0) | ||
| 1173 | && mctx->input.valid_len < mctx->input.len)) | ||
| 1174 | { | ||
| 1175 | err = extend_buffers (mctx); | ||
| 1176 | if (BE (err != REG_NOERROR, 0)) | ||
| 1177 | { | ||
| 1178 | assert (err == REG_ESPACE); | ||
| 1179 | return REG_ERROR; | ||
| 1180 | } | ||
| 1181 | } | ||
| 1182 | |||
| 1183 | cur_state = transit_state (&err, mctx, cur_state); | ||
| 1184 | if (mctx->state_log != NULL) | ||
| 1185 | cur_state = merge_state_with_log (&err, mctx, cur_state); | ||
| 1186 | |||
| 1187 | if (cur_state == NULL) | ||
| 1188 | { | ||
| 1189 | /* Reached the invalid state or an error. Try to recover a valid | ||
| 1190 | state using the state log, if available and if we have not | ||
| 1191 | already found a valid (even if not the longest) match. */ | ||
| 1192 | if (BE (err != REG_NOERROR, 0)) | ||
| 1193 | return REG_ERROR; | ||
| 1194 | |||
| 1195 | if (mctx->state_log == NULL | ||
| 1196 | || (match && !fl_longest_match) | ||
| 1197 | || (cur_state = find_recover_state (&err, mctx)) == NULL) | ||
| 1198 | break; | ||
| 1199 | } | ||
| 1200 | |||
| 1201 | if (BE (at_init_state, 0)) | ||
| 1202 | { | ||
| 1203 | if (old_state == cur_state) | ||
| 1204 | next_start_idx = next_char_idx; | ||
| 1205 | else | ||
| 1206 | at_init_state = false; | ||
| 1207 | } | ||
| 1208 | |||
| 1209 | if (cur_state->halt) | ||
| 1210 | { | ||
| 1211 | /* Reached a halt state. | ||
| 1212 | Check the halt state can satisfy the current context. */ | ||
| 1213 | if (!cur_state->has_constraint | ||
| 1214 | || check_halt_state_context (mctx, cur_state, | ||
| 1215 | re_string_cur_idx (&mctx->input))) | ||
| 1216 | { | ||
| 1217 | /* We found an appropriate halt state. */ | ||
| 1218 | match_last = re_string_cur_idx (&mctx->input); | ||
| 1219 | match = 1; | ||
| 1220 | |||
| 1221 | /* We found a match, do not modify match_first below. */ | ||
| 1222 | p_match_first = NULL; | ||
| 1223 | if (!fl_longest_match) | ||
| 1224 | break; | ||
| 1225 | } | ||
| 1226 | } | ||
| 1227 | } | ||
| 1228 | |||
| 1229 | if (p_match_first) | ||
| 1230 | *p_match_first += next_start_idx; | ||
| 1231 | |||
| 1232 | return match_last; | ||
| 1233 | } | ||
| 1234 | |||
| 1235 | /* Check NODE match the current context. */ | ||
| 1236 | |||
| 1237 | static bool | ||
| 1238 | internal_function | ||
| 1239 | check_halt_node_context (const re_dfa_t *dfa, Idx node, unsigned int context) | ||
| 1240 | { | ||
| 1241 | re_token_type_t type = dfa->nodes[node].type; | ||
| 1242 | unsigned int constraint = dfa->nodes[node].constraint; | ||
| 1243 | if (type != END_OF_RE) | ||
| 1244 | return false; | ||
| 1245 | if (!constraint) | ||
| 1246 | return true; | ||
| 1247 | if (NOT_SATISFY_NEXT_CONSTRAINT (constraint, context)) | ||
| 1248 | return false; | ||
| 1249 | return true; | ||
| 1250 | } | ||
| 1251 | |||
| 1252 | /* Check the halt state STATE match the current context. | ||
| 1253 | Return 0 if not match, if the node, STATE has, is a halt node and | ||
| 1254 | match the context, return the node. */ | ||
| 1255 | |||
| 1256 | static Idx | ||
| 1257 | internal_function | ||
| 1258 | check_halt_state_context (const re_match_context_t *mctx, | ||
| 1259 | const re_dfastate_t *state, Idx idx) | ||
| 1260 | { | ||
| 1261 | Idx i; | ||
| 1262 | unsigned int context; | ||
| 1263 | #ifdef DEBUG | ||
| 1264 | assert (state->halt); | ||
| 1265 | #endif | ||
| 1266 | context = re_string_context_at (&mctx->input, idx, mctx->eflags); | ||
| 1267 | for (i = 0; i < state->nodes.nelem; ++i) | ||
| 1268 | if (check_halt_node_context (mctx->dfa, state->nodes.elems[i], context)) | ||
| 1269 | return state->nodes.elems[i]; | ||
| 1270 | return 0; | ||
| 1271 | } | ||
| 1272 | |||
| 1273 | /* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA | ||
| 1274 | corresponding to the DFA). | ||
| 1275 | Return the destination node, and update EPS_VIA_NODES; | ||
| 1276 | return REG_MISSING in case of errors. */ | ||
| 1277 | |||
| 1278 | static Idx | ||
| 1279 | internal_function | ||
| 1280 | proceed_next_node (const re_match_context_t *mctx, Idx nregs, regmatch_t *regs, | ||
| 1281 | Idx *pidx, Idx node, re_node_set *eps_via_nodes, | ||
| 1282 | struct re_fail_stack_t *fs) | ||
| 1283 | { | ||
| 1284 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1285 | Idx i; | ||
| 1286 | bool ok; | ||
| 1287 | if (IS_EPSILON_NODE (dfa->nodes[node].type)) | ||
| 1288 | { | ||
| 1289 | re_node_set *cur_nodes = &mctx->state_log[*pidx]->nodes; | ||
| 1290 | re_node_set *edests = &dfa->edests[node]; | ||
| 1291 | Idx dest_node; | ||
| 1292 | ok = re_node_set_insert (eps_via_nodes, node); | ||
| 1293 | if (BE (! ok, 0)) | ||
| 1294 | return REG_ERROR; | ||
| 1295 | /* Pick up a valid destination, or return REG_MISSING if none | ||
| 1296 | is found. */ | ||
| 1297 | for (dest_node = REG_MISSING, i = 0; i < edests->nelem; ++i) | ||
| 1298 | { | ||
| 1299 | Idx candidate = edests->elems[i]; | ||
| 1300 | if (!re_node_set_contains (cur_nodes, candidate)) | ||
| 1301 | continue; | ||
| 1302 | if (dest_node == REG_MISSING) | ||
| 1303 | dest_node = candidate; | ||
| 1304 | |||
| 1305 | else | ||
| 1306 | { | ||
| 1307 | /* In order to avoid infinite loop like "(a*)*", return the second | ||
| 1308 | epsilon-transition if the first was already considered. */ | ||
| 1309 | if (re_node_set_contains (eps_via_nodes, dest_node)) | ||
| 1310 | return candidate; | ||
| 1311 | |||
| 1312 | /* Otherwise, push the second epsilon-transition on the fail stack. */ | ||
| 1313 | else if (fs != NULL | ||
| 1314 | && push_fail_stack (fs, *pidx, candidate, nregs, regs, | ||
| 1315 | eps_via_nodes)) | ||
| 1316 | return REG_ERROR; | ||
| 1317 | |||
| 1318 | /* We know we are going to exit. */ | ||
| 1319 | break; | ||
| 1320 | } | ||
| 1321 | } | ||
| 1322 | return dest_node; | ||
| 1323 | } | ||
| 1324 | else | ||
| 1325 | { | ||
| 1326 | Idx naccepted = 0; | ||
| 1327 | re_token_type_t type = dfa->nodes[node].type; | ||
| 1328 | |||
| 1329 | #ifdef RE_ENABLE_I18N | ||
| 1330 | if (dfa->nodes[node].accept_mb) | ||
| 1331 | naccepted = check_node_accept_bytes (dfa, node, &mctx->input, *pidx); | ||
| 1332 | else | ||
| 1333 | #endif /* RE_ENABLE_I18N */ | ||
| 1334 | if (type == OP_BACK_REF) | ||
| 1335 | { | ||
| 1336 | Idx subexp_idx = dfa->nodes[node].opr.idx + 1; | ||
| 1337 | naccepted = regs[subexp_idx].rm_eo - regs[subexp_idx].rm_so; | ||
| 1338 | if (fs != NULL) | ||
| 1339 | { | ||
| 1340 | if (regs[subexp_idx].rm_so == -1 || regs[subexp_idx].rm_eo == -1) | ||
| 1341 | return REG_MISSING; | ||
| 1342 | else if (naccepted) | ||
| 1343 | { | ||
| 1344 | char *buf = (char *) re_string_get_buffer (&mctx->input); | ||
| 1345 | if (memcmp (buf + regs[subexp_idx].rm_so, buf + *pidx, | ||
| 1346 | naccepted) != 0) | ||
| 1347 | return REG_MISSING; | ||
| 1348 | } | ||
| 1349 | } | ||
| 1350 | |||
| 1351 | if (naccepted == 0) | ||
| 1352 | { | ||
| 1353 | Idx dest_node; | ||
| 1354 | ok = re_node_set_insert (eps_via_nodes, node); | ||
| 1355 | if (BE (! ok, 0)) | ||
| 1356 | return REG_ERROR; | ||
| 1357 | dest_node = dfa->edests[node].elems[0]; | ||
| 1358 | if (re_node_set_contains (&mctx->state_log[*pidx]->nodes, | ||
| 1359 | dest_node)) | ||
| 1360 | return dest_node; | ||
| 1361 | } | ||
| 1362 | } | ||
| 1363 | |||
| 1364 | if (naccepted != 0 | ||
| 1365 | || check_node_accept (mctx, dfa->nodes + node, *pidx)) | ||
| 1366 | { | ||
| 1367 | Idx dest_node = dfa->nexts[node]; | ||
| 1368 | *pidx = (naccepted == 0) ? *pidx + 1 : *pidx + naccepted; | ||
| 1369 | if (fs && (*pidx > mctx->match_last || mctx->state_log[*pidx] == NULL | ||
| 1370 | || !re_node_set_contains (&mctx->state_log[*pidx]->nodes, | ||
| 1371 | dest_node))) | ||
| 1372 | return REG_MISSING; | ||
| 1373 | re_node_set_empty (eps_via_nodes); | ||
| 1374 | return dest_node; | ||
| 1375 | } | ||
| 1376 | } | ||
| 1377 | return REG_MISSING; | ||
| 1378 | } | ||
| 1379 | |||
| 1380 | static reg_errcode_t | ||
| 1381 | internal_function | ||
| 1382 | push_fail_stack (struct re_fail_stack_t *fs, Idx str_idx, Idx dest_node, | ||
| 1383 | Idx nregs, regmatch_t *regs, re_node_set *eps_via_nodes) | ||
| 1384 | { | ||
| 1385 | reg_errcode_t err; | ||
| 1386 | Idx num = fs->num++; | ||
| 1387 | if (fs->num == fs->alloc) | ||
| 1388 | { | ||
| 1389 | struct re_fail_stack_ent_t *new_array; | ||
| 1390 | new_array = realloc (fs->stack, (sizeof (struct re_fail_stack_ent_t) | ||
| 1391 | * fs->alloc * 2)); | ||
| 1392 | if (new_array == NULL) | ||
| 1393 | return REG_ESPACE; | ||
| 1394 | fs->alloc *= 2; | ||
| 1395 | fs->stack = new_array; | ||
| 1396 | } | ||
| 1397 | fs->stack[num].idx = str_idx; | ||
| 1398 | fs->stack[num].node = dest_node; | ||
| 1399 | fs->stack[num].regs = re_malloc (regmatch_t, nregs); | ||
| 1400 | if (fs->stack[num].regs == NULL) | ||
| 1401 | return REG_ESPACE; | ||
| 1402 | memcpy (fs->stack[num].regs, regs, sizeof (regmatch_t) * nregs); | ||
| 1403 | err = re_node_set_init_copy (&fs->stack[num].eps_via_nodes, eps_via_nodes); | ||
| 1404 | return err; | ||
| 1405 | } | ||
| 1406 | |||
| 1407 | static Idx | ||
| 1408 | internal_function | ||
| 1409 | pop_fail_stack (struct re_fail_stack_t *fs, Idx *pidx, Idx nregs, | ||
| 1410 | regmatch_t *regs, re_node_set *eps_via_nodes) | ||
| 1411 | { | ||
| 1412 | Idx num = --fs->num; | ||
| 1413 | assert (REG_VALID_INDEX (num)); | ||
| 1414 | *pidx = fs->stack[num].idx; | ||
| 1415 | memcpy (regs, fs->stack[num].regs, sizeof (regmatch_t) * nregs); | ||
| 1416 | re_node_set_free (eps_via_nodes); | ||
| 1417 | re_free (fs->stack[num].regs); | ||
| 1418 | *eps_via_nodes = fs->stack[num].eps_via_nodes; | ||
| 1419 | return fs->stack[num].node; | ||
| 1420 | } | ||
| 1421 | |||
| 1422 | /* Set the positions where the subexpressions are starts/ends to registers | ||
| 1423 | PMATCH. | ||
| 1424 | Note: We assume that pmatch[0] is already set, and | ||
| 1425 | pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */ | ||
| 1426 | |||
| 1427 | static reg_errcode_t | ||
| 1428 | internal_function | ||
| 1429 | set_regs (const regex_t *preg, const re_match_context_t *mctx, size_t nmatch, | ||
| 1430 | regmatch_t *pmatch, bool fl_backtrack) | ||
| 1431 | { | ||
| 1432 | const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer; | ||
| 1433 | Idx idx, cur_node; | ||
| 1434 | re_node_set eps_via_nodes; | ||
| 1435 | struct re_fail_stack_t *fs; | ||
| 1436 | struct re_fail_stack_t fs_body = { 0, 2, NULL }; | ||
| 1437 | regmatch_t *prev_idx_match; | ||
| 1438 | bool prev_idx_match_malloced = false; | ||
| 1439 | |||
| 1440 | #ifdef DEBUG | ||
| 1441 | assert (nmatch > 1); | ||
| 1442 | assert (mctx->state_log != NULL); | ||
| 1443 | #endif | ||
| 1444 | if (fl_backtrack) | ||
| 1445 | { | ||
| 1446 | fs = &fs_body; | ||
| 1447 | fs->stack = re_malloc (struct re_fail_stack_ent_t, fs->alloc); | ||
| 1448 | if (fs->stack == NULL) | ||
| 1449 | return REG_ESPACE; | ||
| 1450 | } | ||
| 1451 | else | ||
| 1452 | fs = NULL; | ||
| 1453 | |||
| 1454 | cur_node = dfa->init_node; | ||
| 1455 | re_node_set_init_empty (&eps_via_nodes); | ||
| 1456 | |||
| 1457 | if (__libc_use_alloca (nmatch * sizeof (regmatch_t))) | ||
| 1458 | prev_idx_match = (regmatch_t *) alloca (nmatch * sizeof (regmatch_t)); | ||
| 1459 | else | ||
| 1460 | { | ||
| 1461 | prev_idx_match = re_malloc (regmatch_t, nmatch); | ||
| 1462 | if (prev_idx_match == NULL) | ||
| 1463 | { | ||
| 1464 | free_fail_stack_return (fs); | ||
| 1465 | return REG_ESPACE; | ||
| 1466 | } | ||
| 1467 | prev_idx_match_malloced = true; | ||
| 1468 | } | ||
| 1469 | memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch); | ||
| 1470 | |||
| 1471 | for (idx = pmatch[0].rm_so; idx <= pmatch[0].rm_eo ;) | ||
| 1472 | { | ||
| 1473 | update_regs (dfa, pmatch, prev_idx_match, cur_node, idx, nmatch); | ||
| 1474 | |||
| 1475 | if (idx == pmatch[0].rm_eo && cur_node == mctx->last_node) | ||
| 1476 | { | ||
| 1477 | Idx reg_idx; | ||
| 1478 | if (fs) | ||
| 1479 | { | ||
| 1480 | for (reg_idx = 0; reg_idx < nmatch; ++reg_idx) | ||
| 1481 | if (pmatch[reg_idx].rm_so > -1 && pmatch[reg_idx].rm_eo == -1) | ||
| 1482 | break; | ||
| 1483 | if (reg_idx == nmatch) | ||
| 1484 | { | ||
| 1485 | re_node_set_free (&eps_via_nodes); | ||
| 1486 | if (prev_idx_match_malloced) | ||
| 1487 | re_free (prev_idx_match); | ||
| 1488 | return free_fail_stack_return (fs); | ||
| 1489 | } | ||
| 1490 | cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch, | ||
| 1491 | &eps_via_nodes); | ||
| 1492 | } | ||
| 1493 | else | ||
| 1494 | { | ||
| 1495 | re_node_set_free (&eps_via_nodes); | ||
| 1496 | if (prev_idx_match_malloced) | ||
| 1497 | re_free (prev_idx_match); | ||
| 1498 | return REG_NOERROR; | ||
| 1499 | } | ||
| 1500 | } | ||
| 1501 | |||
| 1502 | /* Proceed to next node. */ | ||
| 1503 | cur_node = proceed_next_node (mctx, nmatch, pmatch, &idx, cur_node, | ||
| 1504 | &eps_via_nodes, fs); | ||
| 1505 | |||
| 1506 | if (BE (! REG_VALID_INDEX (cur_node), 0)) | ||
| 1507 | { | ||
| 1508 | if (BE (cur_node == REG_ERROR, 0)) | ||
| 1509 | { | ||
| 1510 | re_node_set_free (&eps_via_nodes); | ||
| 1511 | if (prev_idx_match_malloced) | ||
| 1512 | re_free (prev_idx_match); | ||
| 1513 | free_fail_stack_return (fs); | ||
| 1514 | return REG_ESPACE; | ||
| 1515 | } | ||
| 1516 | if (fs) | ||
| 1517 | cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch, | ||
| 1518 | &eps_via_nodes); | ||
| 1519 | else | ||
| 1520 | { | ||
| 1521 | re_node_set_free (&eps_via_nodes); | ||
| 1522 | if (prev_idx_match_malloced) | ||
| 1523 | re_free (prev_idx_match); | ||
| 1524 | return REG_NOMATCH; | ||
| 1525 | } | ||
| 1526 | } | ||
| 1527 | } | ||
| 1528 | re_node_set_free (&eps_via_nodes); | ||
| 1529 | if (prev_idx_match_malloced) | ||
| 1530 | re_free (prev_idx_match); | ||
| 1531 | return free_fail_stack_return (fs); | ||
| 1532 | } | ||
| 1533 | |||
| 1534 | static reg_errcode_t | ||
| 1535 | internal_function | ||
| 1536 | free_fail_stack_return (struct re_fail_stack_t *fs) | ||
| 1537 | { | ||
| 1538 | if (fs) | ||
| 1539 | { | ||
| 1540 | Idx fs_idx; | ||
| 1541 | for (fs_idx = 0; fs_idx < fs->num; ++fs_idx) | ||
| 1542 | { | ||
| 1543 | re_node_set_free (&fs->stack[fs_idx].eps_via_nodes); | ||
| 1544 | re_free (fs->stack[fs_idx].regs); | ||
| 1545 | } | ||
| 1546 | re_free (fs->stack); | ||
| 1547 | } | ||
| 1548 | return REG_NOERROR; | ||
| 1549 | } | ||
| 1550 | |||
| 1551 | static void | ||
| 1552 | internal_function | ||
| 1553 | update_regs (const re_dfa_t *dfa, regmatch_t *pmatch, | ||
| 1554 | regmatch_t *prev_idx_match, Idx cur_node, Idx cur_idx, Idx nmatch) | ||
| 1555 | { | ||
| 1556 | int type = dfa->nodes[cur_node].type; | ||
| 1557 | if (type == OP_OPEN_SUBEXP) | ||
| 1558 | { | ||
| 1559 | Idx reg_num = dfa->nodes[cur_node].opr.idx + 1; | ||
| 1560 | |||
| 1561 | /* We are at the first node of this sub expression. */ | ||
| 1562 | if (reg_num < nmatch) | ||
| 1563 | { | ||
| 1564 | pmatch[reg_num].rm_so = cur_idx; | ||
| 1565 | pmatch[reg_num].rm_eo = -1; | ||
| 1566 | } | ||
| 1567 | } | ||
| 1568 | else if (type == OP_CLOSE_SUBEXP) | ||
| 1569 | { | ||
| 1570 | Idx reg_num = dfa->nodes[cur_node].opr.idx + 1; | ||
| 1571 | if (reg_num < nmatch) | ||
| 1572 | { | ||
| 1573 | /* We are at the last node of this sub expression. */ | ||
| 1574 | if (pmatch[reg_num].rm_so < cur_idx) | ||
| 1575 | { | ||
| 1576 | pmatch[reg_num].rm_eo = cur_idx; | ||
| 1577 | /* This is a non-empty match or we are not inside an optional | ||
| 1578 | subexpression. Accept this right away. */ | ||
| 1579 | memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch); | ||
| 1580 | } | ||
| 1581 | else | ||
| 1582 | { | ||
| 1583 | if (dfa->nodes[cur_node].opt_subexp | ||
| 1584 | && prev_idx_match[reg_num].rm_so != -1) | ||
| 1585 | /* We transited through an empty match for an optional | ||
| 1586 | subexpression, like (a?)*, and this is not the subexp's | ||
| 1587 | first match. Copy back the old content of the registers | ||
| 1588 | so that matches of an inner subexpression are undone as | ||
| 1589 | well, like in ((a?))*. */ | ||
| 1590 | memcpy (pmatch, prev_idx_match, sizeof (regmatch_t) * nmatch); | ||
| 1591 | else | ||
| 1592 | /* We completed a subexpression, but it may be part of | ||
| 1593 | an optional one, so do not update PREV_IDX_MATCH. */ | ||
| 1594 | pmatch[reg_num].rm_eo = cur_idx; | ||
| 1595 | } | ||
| 1596 | } | ||
| 1597 | } | ||
| 1598 | } | ||
| 1599 | |||
| 1600 | /* This function checks the STATE_LOG from the SCTX->last_str_idx to 0 | ||
| 1601 | and sift the nodes in each states according to the following rules. | ||
| 1602 | Updated state_log will be wrote to STATE_LOG. | ||
| 1603 | |||
| 1604 | Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if... | ||
| 1605 | 1. When STR_IDX == MATCH_LAST(the last index in the state_log): | ||
| 1606 | If `a' isn't the LAST_NODE and `a' can't epsilon transit to | ||
| 1607 | the LAST_NODE, we throw away the node `a'. | ||
| 1608 | 2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts | ||
| 1609 | string `s' and transit to `b': | ||
| 1610 | i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw | ||
| 1611 | away the node `a'. | ||
| 1612 | ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is | ||
| 1613 | thrown away, we throw away the node `a'. | ||
| 1614 | 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b': | ||
| 1615 | i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the | ||
| 1616 | node `a'. | ||
| 1617 | ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away, | ||
| 1618 | we throw away the node `a'. */ | ||
| 1619 | |||
| 1620 | #define STATE_NODE_CONTAINS(state,node) \ | ||
| 1621 | ((state) != NULL && re_node_set_contains (&(state)->nodes, node)) | ||
| 1622 | |||
| 1623 | static reg_errcode_t | ||
| 1624 | internal_function | ||
| 1625 | sift_states_backward (const re_match_context_t *mctx, re_sift_context_t *sctx) | ||
| 1626 | { | ||
| 1627 | reg_errcode_t err; | ||
| 1628 | int null_cnt = 0; | ||
| 1629 | Idx str_idx = sctx->last_str_idx; | ||
| 1630 | re_node_set cur_dest; | ||
| 1631 | |||
| 1632 | #ifdef DEBUG | ||
| 1633 | assert (mctx->state_log != NULL && mctx->state_log[str_idx] != NULL); | ||
| 1634 | #endif | ||
| 1635 | |||
| 1636 | /* Build sifted state_log[str_idx]. It has the nodes which can epsilon | ||
| 1637 | transit to the last_node and the last_node itself. */ | ||
| 1638 | err = re_node_set_init_1 (&cur_dest, sctx->last_node); | ||
| 1639 | if (BE (err != REG_NOERROR, 0)) | ||
| 1640 | return err; | ||
| 1641 | err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest); | ||
| 1642 | if (BE (err != REG_NOERROR, 0)) | ||
| 1643 | goto free_return; | ||
| 1644 | |||
| 1645 | /* Then check each states in the state_log. */ | ||
| 1646 | while (str_idx > 0) | ||
| 1647 | { | ||
| 1648 | /* Update counters. */ | ||
| 1649 | null_cnt = (sctx->sifted_states[str_idx] == NULL) ? null_cnt + 1 : 0; | ||
| 1650 | if (null_cnt > mctx->max_mb_elem_len) | ||
| 1651 | { | ||
| 1652 | memset (sctx->sifted_states, '\0', | ||
| 1653 | sizeof (re_dfastate_t *) * str_idx); | ||
| 1654 | re_node_set_free (&cur_dest); | ||
| 1655 | return REG_NOERROR; | ||
| 1656 | } | ||
| 1657 | re_node_set_empty (&cur_dest); | ||
| 1658 | --str_idx; | ||
| 1659 | |||
| 1660 | if (mctx->state_log[str_idx]) | ||
| 1661 | { | ||
| 1662 | err = build_sifted_states (mctx, sctx, str_idx, &cur_dest); | ||
| 1663 | if (BE (err != REG_NOERROR, 0)) | ||
| 1664 | goto free_return; | ||
| 1665 | } | ||
| 1666 | |||
| 1667 | /* Add all the nodes which satisfy the following conditions: | ||
| 1668 | - It can epsilon transit to a node in CUR_DEST. | ||
| 1669 | - It is in CUR_SRC. | ||
| 1670 | And update state_log. */ | ||
| 1671 | err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest); | ||
| 1672 | if (BE (err != REG_NOERROR, 0)) | ||
| 1673 | goto free_return; | ||
| 1674 | } | ||
| 1675 | err = REG_NOERROR; | ||
| 1676 | free_return: | ||
| 1677 | re_node_set_free (&cur_dest); | ||
| 1678 | return err; | ||
| 1679 | } | ||
| 1680 | |||
| 1681 | static reg_errcode_t | ||
| 1682 | internal_function | ||
| 1683 | build_sifted_states (const re_match_context_t *mctx, re_sift_context_t *sctx, | ||
| 1684 | Idx str_idx, re_node_set *cur_dest) | ||
| 1685 | { | ||
| 1686 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1687 | const re_node_set *cur_src = &mctx->state_log[str_idx]->non_eps_nodes; | ||
| 1688 | Idx i; | ||
| 1689 | |||
| 1690 | /* Then build the next sifted state. | ||
| 1691 | We build the next sifted state on `cur_dest', and update | ||
| 1692 | `sifted_states[str_idx]' with `cur_dest'. | ||
| 1693 | Note: | ||
| 1694 | `cur_dest' is the sifted state from `state_log[str_idx + 1]'. | ||
| 1695 | `cur_src' points the node_set of the old `state_log[str_idx]' | ||
| 1696 | (with the epsilon nodes pre-filtered out). */ | ||
| 1697 | for (i = 0; i < cur_src->nelem; i++) | ||
| 1698 | { | ||
| 1699 | Idx prev_node = cur_src->elems[i]; | ||
| 1700 | int naccepted = 0; | ||
| 1701 | bool ok; | ||
| 1702 | |||
| 1703 | #ifdef DEBUG | ||
| 1704 | re_token_type_t type = dfa->nodes[prev_node].type; | ||
| 1705 | assert (!IS_EPSILON_NODE (type)); | ||
| 1706 | #endif | ||
| 1707 | #ifdef RE_ENABLE_I18N | ||
| 1708 | /* If the node may accept `multi byte'. */ | ||
| 1709 | if (dfa->nodes[prev_node].accept_mb) | ||
| 1710 | naccepted = sift_states_iter_mb (mctx, sctx, prev_node, | ||
| 1711 | str_idx, sctx->last_str_idx); | ||
| 1712 | #endif /* RE_ENABLE_I18N */ | ||
| 1713 | |||
| 1714 | /* We don't check backreferences here. | ||
| 1715 | See update_cur_sifted_state(). */ | ||
| 1716 | if (!naccepted | ||
| 1717 | && check_node_accept (mctx, dfa->nodes + prev_node, str_idx) | ||
| 1718 | && STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + 1], | ||
| 1719 | dfa->nexts[prev_node])) | ||
| 1720 | naccepted = 1; | ||
| 1721 | |||
| 1722 | if (naccepted == 0) | ||
| 1723 | continue; | ||
| 1724 | |||
| 1725 | if (sctx->limits.nelem) | ||
| 1726 | { | ||
| 1727 | Idx to_idx = str_idx + naccepted; | ||
| 1728 | if (check_dst_limits (mctx, &sctx->limits, | ||
| 1729 | dfa->nexts[prev_node], to_idx, | ||
| 1730 | prev_node, str_idx)) | ||
| 1731 | continue; | ||
| 1732 | } | ||
| 1733 | ok = re_node_set_insert (cur_dest, prev_node); | ||
| 1734 | if (BE (! ok, 0)) | ||
| 1735 | return REG_ESPACE; | ||
| 1736 | } | ||
| 1737 | |||
| 1738 | return REG_NOERROR; | ||
| 1739 | } | ||
| 1740 | |||
| 1741 | /* Helper functions. */ | ||
| 1742 | |||
| 1743 | static reg_errcode_t | ||
| 1744 | internal_function | ||
| 1745 | clean_state_log_if_needed (re_match_context_t *mctx, Idx next_state_log_idx) | ||
| 1746 | { | ||
| 1747 | Idx top = mctx->state_log_top; | ||
| 1748 | |||
| 1749 | if (next_state_log_idx >= mctx->input.bufs_len | ||
| 1750 | || (next_state_log_idx >= mctx->input.valid_len | ||
| 1751 | && mctx->input.valid_len < mctx->input.len)) | ||
| 1752 | { | ||
| 1753 | reg_errcode_t err; | ||
| 1754 | err = extend_buffers (mctx); | ||
| 1755 | if (BE (err != REG_NOERROR, 0)) | ||
| 1756 | return err; | ||
| 1757 | } | ||
| 1758 | |||
| 1759 | if (top < next_state_log_idx) | ||
| 1760 | { | ||
| 1761 | memset (mctx->state_log + top + 1, '\0', | ||
| 1762 | sizeof (re_dfastate_t *) * (next_state_log_idx - top)); | ||
| 1763 | mctx->state_log_top = next_state_log_idx; | ||
| 1764 | } | ||
| 1765 | return REG_NOERROR; | ||
| 1766 | } | ||
| 1767 | |||
| 1768 | static reg_errcode_t | ||
| 1769 | internal_function | ||
| 1770 | merge_state_array (const re_dfa_t *dfa, re_dfastate_t **dst, | ||
| 1771 | re_dfastate_t **src, Idx num) | ||
| 1772 | { | ||
| 1773 | Idx st_idx; | ||
| 1774 | reg_errcode_t err; | ||
| 1775 | for (st_idx = 0; st_idx < num; ++st_idx) | ||
| 1776 | { | ||
| 1777 | if (dst[st_idx] == NULL) | ||
| 1778 | dst[st_idx] = src[st_idx]; | ||
| 1779 | else if (src[st_idx] != NULL) | ||
| 1780 | { | ||
| 1781 | re_node_set merged_set; | ||
| 1782 | err = re_node_set_init_union (&merged_set, &dst[st_idx]->nodes, | ||
| 1783 | &src[st_idx]->nodes); | ||
| 1784 | if (BE (err != REG_NOERROR, 0)) | ||
| 1785 | return err; | ||
| 1786 | dst[st_idx] = re_acquire_state (&err, dfa, &merged_set); | ||
| 1787 | re_node_set_free (&merged_set); | ||
| 1788 | if (BE (err != REG_NOERROR, 0)) | ||
| 1789 | return err; | ||
| 1790 | } | ||
| 1791 | } | ||
| 1792 | return REG_NOERROR; | ||
| 1793 | } | ||
| 1794 | |||
| 1795 | static reg_errcode_t | ||
| 1796 | internal_function | ||
| 1797 | update_cur_sifted_state (const re_match_context_t *mctx, | ||
| 1798 | re_sift_context_t *sctx, Idx str_idx, | ||
| 1799 | re_node_set *dest_nodes) | ||
| 1800 | { | ||
| 1801 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1802 | reg_errcode_t err = REG_NOERROR; | ||
| 1803 | const re_node_set *candidates; | ||
| 1804 | candidates = ((mctx->state_log[str_idx] == NULL) ? NULL | ||
| 1805 | : &mctx->state_log[str_idx]->nodes); | ||
| 1806 | |||
| 1807 | if (dest_nodes->nelem == 0) | ||
| 1808 | sctx->sifted_states[str_idx] = NULL; | ||
| 1809 | else | ||
| 1810 | { | ||
| 1811 | if (candidates) | ||
| 1812 | { | ||
| 1813 | /* At first, add the nodes which can epsilon transit to a node in | ||
| 1814 | DEST_NODE. */ | ||
| 1815 | err = add_epsilon_src_nodes (dfa, dest_nodes, candidates); | ||
| 1816 | if (BE (err != REG_NOERROR, 0)) | ||
| 1817 | return err; | ||
| 1818 | |||
| 1819 | /* Then, check the limitations in the current sift_context. */ | ||
| 1820 | if (sctx->limits.nelem) | ||
| 1821 | { | ||
| 1822 | err = check_subexp_limits (dfa, dest_nodes, candidates, &sctx->limits, | ||
| 1823 | mctx->bkref_ents, str_idx); | ||
| 1824 | if (BE (err != REG_NOERROR, 0)) | ||
| 1825 | return err; | ||
| 1826 | } | ||
| 1827 | } | ||
| 1828 | |||
| 1829 | sctx->sifted_states[str_idx] = re_acquire_state (&err, dfa, dest_nodes); | ||
| 1830 | if (BE (err != REG_NOERROR, 0)) | ||
| 1831 | return err; | ||
| 1832 | } | ||
| 1833 | |||
| 1834 | if (candidates && mctx->state_log[str_idx]->has_backref) | ||
| 1835 | { | ||
| 1836 | err = sift_states_bkref (mctx, sctx, str_idx, candidates); | ||
| 1837 | if (BE (err != REG_NOERROR, 0)) | ||
| 1838 | return err; | ||
| 1839 | } | ||
| 1840 | return REG_NOERROR; | ||
| 1841 | } | ||
| 1842 | |||
| 1843 | static reg_errcode_t | ||
| 1844 | internal_function | ||
| 1845 | add_epsilon_src_nodes (const re_dfa_t *dfa, re_node_set *dest_nodes, | ||
| 1846 | const re_node_set *candidates) | ||
| 1847 | { | ||
| 1848 | reg_errcode_t err = REG_NOERROR; | ||
| 1849 | Idx i; | ||
| 1850 | |||
| 1851 | re_dfastate_t *state = re_acquire_state (&err, dfa, dest_nodes); | ||
| 1852 | if (BE (err != REG_NOERROR, 0)) | ||
| 1853 | return err; | ||
| 1854 | |||
| 1855 | if (!state->inveclosure.alloc) | ||
| 1856 | { | ||
| 1857 | err = re_node_set_alloc (&state->inveclosure, dest_nodes->nelem); | ||
| 1858 | if (BE (err != REG_NOERROR, 0)) | ||
| 1859 | return REG_ESPACE; | ||
| 1860 | for (i = 0; i < dest_nodes->nelem; i++) | ||
| 1861 | re_node_set_merge (&state->inveclosure, | ||
| 1862 | dfa->inveclosures + dest_nodes->elems[i]); | ||
| 1863 | } | ||
| 1864 | return re_node_set_add_intersect (dest_nodes, candidates, | ||
| 1865 | &state->inveclosure); | ||
| 1866 | } | ||
| 1867 | |||
| 1868 | static reg_errcode_t | ||
| 1869 | internal_function | ||
| 1870 | sub_epsilon_src_nodes (const re_dfa_t *dfa, Idx node, re_node_set *dest_nodes, | ||
| 1871 | const re_node_set *candidates) | ||
| 1872 | { | ||
| 1873 | Idx ecl_idx; | ||
| 1874 | reg_errcode_t err; | ||
| 1875 | re_node_set *inv_eclosure = dfa->inveclosures + node; | ||
| 1876 | re_node_set except_nodes; | ||
| 1877 | re_node_set_init_empty (&except_nodes); | ||
| 1878 | for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx) | ||
| 1879 | { | ||
| 1880 | Idx cur_node = inv_eclosure->elems[ecl_idx]; | ||
| 1881 | if (cur_node == node) | ||
| 1882 | continue; | ||
| 1883 | if (IS_EPSILON_NODE (dfa->nodes[cur_node].type)) | ||
| 1884 | { | ||
| 1885 | Idx edst1 = dfa->edests[cur_node].elems[0]; | ||
| 1886 | Idx edst2 = ((dfa->edests[cur_node].nelem > 1) | ||
| 1887 | ? dfa->edests[cur_node].elems[1] : REG_MISSING); | ||
| 1888 | if ((!re_node_set_contains (inv_eclosure, edst1) | ||
| 1889 | && re_node_set_contains (dest_nodes, edst1)) | ||
| 1890 | || (REG_VALID_NONZERO_INDEX (edst2) | ||
| 1891 | && !re_node_set_contains (inv_eclosure, edst2) | ||
| 1892 | && re_node_set_contains (dest_nodes, edst2))) | ||
| 1893 | { | ||
| 1894 | err = re_node_set_add_intersect (&except_nodes, candidates, | ||
| 1895 | dfa->inveclosures + cur_node); | ||
| 1896 | if (BE (err != REG_NOERROR, 0)) | ||
| 1897 | { | ||
| 1898 | re_node_set_free (&except_nodes); | ||
| 1899 | return err; | ||
| 1900 | } | ||
| 1901 | } | ||
| 1902 | } | ||
| 1903 | } | ||
| 1904 | for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx) | ||
| 1905 | { | ||
| 1906 | Idx cur_node = inv_eclosure->elems[ecl_idx]; | ||
| 1907 | if (!re_node_set_contains (&except_nodes, cur_node)) | ||
| 1908 | { | ||
| 1909 | Idx idx = re_node_set_contains (dest_nodes, cur_node) - 1; | ||
| 1910 | re_node_set_remove_at (dest_nodes, idx); | ||
| 1911 | } | ||
| 1912 | } | ||
| 1913 | re_node_set_free (&except_nodes); | ||
| 1914 | return REG_NOERROR; | ||
| 1915 | } | ||
| 1916 | |||
| 1917 | static bool | ||
| 1918 | internal_function | ||
| 1919 | check_dst_limits (const re_match_context_t *mctx, const re_node_set *limits, | ||
| 1920 | Idx dst_node, Idx dst_idx, Idx src_node, Idx src_idx) | ||
| 1921 | { | ||
| 1922 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1923 | Idx lim_idx, src_pos, dst_pos; | ||
| 1924 | |||
| 1925 | Idx dst_bkref_idx = search_cur_bkref_entry (mctx, dst_idx); | ||
| 1926 | Idx src_bkref_idx = search_cur_bkref_entry (mctx, src_idx); | ||
| 1927 | for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx) | ||
| 1928 | { | ||
| 1929 | Idx subexp_idx; | ||
| 1930 | struct re_backref_cache_entry *ent; | ||
| 1931 | ent = mctx->bkref_ents + limits->elems[lim_idx]; | ||
| 1932 | subexp_idx = dfa->nodes[ent->node].opr.idx; | ||
| 1933 | |||
| 1934 | dst_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx], | ||
| 1935 | subexp_idx, dst_node, dst_idx, | ||
| 1936 | dst_bkref_idx); | ||
| 1937 | src_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx], | ||
| 1938 | subexp_idx, src_node, src_idx, | ||
| 1939 | src_bkref_idx); | ||
| 1940 | |||
| 1941 | /* In case of: | ||
| 1942 | <src> <dst> ( <subexp> ) | ||
| 1943 | ( <subexp> ) <src> <dst> | ||
| 1944 | ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */ | ||
| 1945 | if (src_pos == dst_pos) | ||
| 1946 | continue; /* This is unrelated limitation. */ | ||
| 1947 | else | ||
| 1948 | return true; | ||
| 1949 | } | ||
| 1950 | return false; | ||
| 1951 | } | ||
| 1952 | |||
| 1953 | static int | ||
| 1954 | internal_function | ||
| 1955 | check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, int boundaries, | ||
| 1956 | Idx subexp_idx, Idx from_node, Idx bkref_idx) | ||
| 1957 | { | ||
| 1958 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 1959 | const re_node_set *eclosures = dfa->eclosures + from_node; | ||
| 1960 | Idx node_idx; | ||
| 1961 | |||
| 1962 | /* Else, we are on the boundary: examine the nodes on the epsilon | ||
| 1963 | closure. */ | ||
| 1964 | for (node_idx = 0; node_idx < eclosures->nelem; ++node_idx) | ||
| 1965 | { | ||
| 1966 | Idx node = eclosures->elems[node_idx]; | ||
| 1967 | switch (dfa->nodes[node].type) | ||
| 1968 | { | ||
| 1969 | case OP_BACK_REF: | ||
| 1970 | if (bkref_idx != REG_MISSING) | ||
| 1971 | { | ||
| 1972 | struct re_backref_cache_entry *ent = mctx->bkref_ents + bkref_idx; | ||
| 1973 | do | ||
| 1974 | { | ||
| 1975 | Idx dst; | ||
| 1976 | int cpos; | ||
| 1977 | |||
| 1978 | if (ent->node != node) | ||
| 1979 | continue; | ||
| 1980 | |||
| 1981 | if (subexp_idx < BITSET_WORD_BITS | ||
| 1982 | && !(ent->eps_reachable_subexps_map | ||
| 1983 | & ((bitset_word_t) 1 << subexp_idx))) | ||
| 1984 | continue; | ||
| 1985 | |||
| 1986 | /* Recurse trying to reach the OP_OPEN_SUBEXP and | ||
| 1987 | OP_CLOSE_SUBEXP cases below. But, if the | ||
| 1988 | destination node is the same node as the source | ||
| 1989 | node, don't recurse because it would cause an | ||
| 1990 | infinite loop: a regex that exhibits this behavior | ||
| 1991 | is ()\1*\1* */ | ||
| 1992 | dst = dfa->edests[node].elems[0]; | ||
| 1993 | if (dst == from_node) | ||
| 1994 | { | ||
| 1995 | if (boundaries & 1) | ||
| 1996 | return -1; | ||
| 1997 | else /* if (boundaries & 2) */ | ||
| 1998 | return 0; | ||
| 1999 | } | ||
| 2000 | |||
| 2001 | cpos = | ||
| 2002 | check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx, | ||
| 2003 | dst, bkref_idx); | ||
| 2004 | if (cpos == -1 /* && (boundaries & 1) */) | ||
| 2005 | return -1; | ||
| 2006 | if (cpos == 0 && (boundaries & 2)) | ||
| 2007 | return 0; | ||
| 2008 | |||
| 2009 | if (subexp_idx < BITSET_WORD_BITS) | ||
| 2010 | ent->eps_reachable_subexps_map | ||
| 2011 | &= ~((bitset_word_t) 1 << subexp_idx); | ||
| 2012 | } | ||
| 2013 | while (ent++->more); | ||
| 2014 | } | ||
| 2015 | break; | ||
| 2016 | |||
| 2017 | case OP_OPEN_SUBEXP: | ||
| 2018 | if ((boundaries & 1) && subexp_idx == dfa->nodes[node].opr.idx) | ||
| 2019 | return -1; | ||
| 2020 | break; | ||
| 2021 | |||
| 2022 | case OP_CLOSE_SUBEXP: | ||
| 2023 | if ((boundaries & 2) && subexp_idx == dfa->nodes[node].opr.idx) | ||
| 2024 | return 0; | ||
| 2025 | break; | ||
| 2026 | |||
| 2027 | default: | ||
| 2028 | break; | ||
| 2029 | } | ||
| 2030 | } | ||
| 2031 | |||
| 2032 | return (boundaries & 2) ? 1 : 0; | ||
| 2033 | } | ||
| 2034 | |||
| 2035 | static int | ||
| 2036 | internal_function | ||
| 2037 | check_dst_limits_calc_pos (const re_match_context_t *mctx, Idx limit, | ||
| 2038 | Idx subexp_idx, Idx from_node, Idx str_idx, | ||
| 2039 | Idx bkref_idx) | ||
| 2040 | { | ||
| 2041 | struct re_backref_cache_entry *lim = mctx->bkref_ents + limit; | ||
| 2042 | int boundaries; | ||
| 2043 | |||
| 2044 | /* If we are outside the range of the subexpression, return -1 or 1. */ | ||
| 2045 | if (str_idx < lim->subexp_from) | ||
| 2046 | return -1; | ||
| 2047 | |||
| 2048 | if (lim->subexp_to < str_idx) | ||
| 2049 | return 1; | ||
| 2050 | |||
| 2051 | /* If we are within the subexpression, return 0. */ | ||
| 2052 | boundaries = (str_idx == lim->subexp_from); | ||
| 2053 | boundaries |= (str_idx == lim->subexp_to) << 1; | ||
| 2054 | if (boundaries == 0) | ||
| 2055 | return 0; | ||
| 2056 | |||
| 2057 | /* Else, examine epsilon closure. */ | ||
| 2058 | return check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx, | ||
| 2059 | from_node, bkref_idx); | ||
| 2060 | } | ||
| 2061 | |||
| 2062 | /* Check the limitations of sub expressions LIMITS, and remove the nodes | ||
| 2063 | which are against limitations from DEST_NODES. */ | ||
| 2064 | |||
| 2065 | static reg_errcode_t | ||
| 2066 | internal_function | ||
| 2067 | check_subexp_limits (const re_dfa_t *dfa, re_node_set *dest_nodes, | ||
| 2068 | const re_node_set *candidates, re_node_set *limits, | ||
| 2069 | struct re_backref_cache_entry *bkref_ents, Idx str_idx) | ||
| 2070 | { | ||
| 2071 | reg_errcode_t err; | ||
| 2072 | Idx node_idx, lim_idx; | ||
| 2073 | |||
| 2074 | for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx) | ||
| 2075 | { | ||
| 2076 | Idx subexp_idx; | ||
| 2077 | struct re_backref_cache_entry *ent; | ||
| 2078 | ent = bkref_ents + limits->elems[lim_idx]; | ||
| 2079 | |||
| 2080 | if (str_idx <= ent->subexp_from || ent->str_idx < str_idx) | ||
| 2081 | continue; /* This is unrelated limitation. */ | ||
| 2082 | |||
| 2083 | subexp_idx = dfa->nodes[ent->node].opr.idx; | ||
| 2084 | if (ent->subexp_to == str_idx) | ||
| 2085 | { | ||
| 2086 | Idx ops_node = REG_MISSING; | ||
| 2087 | Idx cls_node = REG_MISSING; | ||
| 2088 | for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx) | ||
| 2089 | { | ||
| 2090 | Idx node = dest_nodes->elems[node_idx]; | ||
| 2091 | re_token_type_t type = dfa->nodes[node].type; | ||
| 2092 | if (type == OP_OPEN_SUBEXP | ||
| 2093 | && subexp_idx == dfa->nodes[node].opr.idx) | ||
| 2094 | ops_node = node; | ||
| 2095 | else if (type == OP_CLOSE_SUBEXP | ||
| 2096 | && subexp_idx == dfa->nodes[node].opr.idx) | ||
| 2097 | cls_node = node; | ||
| 2098 | } | ||
| 2099 | |||
| 2100 | /* Check the limitation of the open subexpression. */ | ||
| 2101 | /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */ | ||
| 2102 | if (REG_VALID_INDEX (ops_node)) | ||
| 2103 | { | ||
| 2104 | err = sub_epsilon_src_nodes (dfa, ops_node, dest_nodes, | ||
| 2105 | candidates); | ||
| 2106 | if (BE (err != REG_NOERROR, 0)) | ||
| 2107 | return err; | ||
| 2108 | } | ||
| 2109 | |||
| 2110 | /* Check the limitation of the close subexpression. */ | ||
| 2111 | if (REG_VALID_INDEX (cls_node)) | ||
| 2112 | for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx) | ||
| 2113 | { | ||
| 2114 | Idx node = dest_nodes->elems[node_idx]; | ||
| 2115 | if (!re_node_set_contains (dfa->inveclosures + node, | ||
| 2116 | cls_node) | ||
| 2117 | && !re_node_set_contains (dfa->eclosures + node, | ||
| 2118 | cls_node)) | ||
| 2119 | { | ||
| 2120 | /* It is against this limitation. | ||
| 2121 | Remove it form the current sifted state. */ | ||
| 2122 | err = sub_epsilon_src_nodes (dfa, node, dest_nodes, | ||
| 2123 | candidates); | ||
| 2124 | if (BE (err != REG_NOERROR, 0)) | ||
| 2125 | return err; | ||
| 2126 | --node_idx; | ||
| 2127 | } | ||
| 2128 | } | ||
| 2129 | } | ||
| 2130 | else /* (ent->subexp_to != str_idx) */ | ||
| 2131 | { | ||
| 2132 | for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx) | ||
| 2133 | { | ||
| 2134 | Idx node = dest_nodes->elems[node_idx]; | ||
| 2135 | re_token_type_t type = dfa->nodes[node].type; | ||
| 2136 | if (type == OP_CLOSE_SUBEXP || type == OP_OPEN_SUBEXP) | ||
| 2137 | { | ||
| 2138 | if (subexp_idx != dfa->nodes[node].opr.idx) | ||
| 2139 | continue; | ||
| 2140 | /* It is against this limitation. | ||
| 2141 | Remove it form the current sifted state. */ | ||
| 2142 | err = sub_epsilon_src_nodes (dfa, node, dest_nodes, | ||
| 2143 | candidates); | ||
| 2144 | if (BE (err != REG_NOERROR, 0)) | ||
| 2145 | return err; | ||
| 2146 | } | ||
| 2147 | } | ||
| 2148 | } | ||
| 2149 | } | ||
| 2150 | return REG_NOERROR; | ||
| 2151 | } | ||
| 2152 | |||
| 2153 | static reg_errcode_t | ||
| 2154 | internal_function | ||
| 2155 | sift_states_bkref (const re_match_context_t *mctx, re_sift_context_t *sctx, | ||
| 2156 | Idx str_idx, const re_node_set *candidates) | ||
| 2157 | { | ||
| 2158 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2159 | reg_errcode_t err; | ||
| 2160 | Idx node_idx, node; | ||
| 2161 | re_sift_context_t local_sctx; | ||
| 2162 | Idx first_idx = search_cur_bkref_entry (mctx, str_idx); | ||
| 2163 | |||
| 2164 | if (first_idx == REG_MISSING) | ||
| 2165 | return REG_NOERROR; | ||
| 2166 | |||
| 2167 | local_sctx.sifted_states = NULL; /* Mark that it hasn't been initialized. */ | ||
| 2168 | |||
| 2169 | for (node_idx = 0; node_idx < candidates->nelem; ++node_idx) | ||
| 2170 | { | ||
| 2171 | Idx enabled_idx; | ||
| 2172 | re_token_type_t type; | ||
| 2173 | struct re_backref_cache_entry *entry; | ||
| 2174 | node = candidates->elems[node_idx]; | ||
| 2175 | type = dfa->nodes[node].type; | ||
| 2176 | /* Avoid infinite loop for the REs like "()\1+". */ | ||
| 2177 | if (node == sctx->last_node && str_idx == sctx->last_str_idx) | ||
| 2178 | continue; | ||
| 2179 | if (type != OP_BACK_REF) | ||
| 2180 | continue; | ||
| 2181 | |||
| 2182 | entry = mctx->bkref_ents + first_idx; | ||
| 2183 | enabled_idx = first_idx; | ||
| 2184 | do | ||
| 2185 | { | ||
| 2186 | Idx subexp_len; | ||
| 2187 | Idx to_idx; | ||
| 2188 | Idx dst_node; | ||
| 2189 | bool ok; | ||
| 2190 | re_dfastate_t *cur_state; | ||
| 2191 | |||
| 2192 | if (entry->node != node) | ||
| 2193 | continue; | ||
| 2194 | subexp_len = entry->subexp_to - entry->subexp_from; | ||
| 2195 | to_idx = str_idx + subexp_len; | ||
| 2196 | dst_node = (subexp_len ? dfa->nexts[node] | ||
| 2197 | : dfa->edests[node].elems[0]); | ||
| 2198 | |||
| 2199 | if (to_idx > sctx->last_str_idx | ||
| 2200 | || sctx->sifted_states[to_idx] == NULL | ||
| 2201 | || !STATE_NODE_CONTAINS (sctx->sifted_states[to_idx], dst_node) | ||
| 2202 | || check_dst_limits (mctx, &sctx->limits, node, | ||
| 2203 | str_idx, dst_node, to_idx)) | ||
| 2204 | continue; | ||
| 2205 | |||
| 2206 | if (local_sctx.sifted_states == NULL) | ||
| 2207 | { | ||
| 2208 | local_sctx = *sctx; | ||
| 2209 | err = re_node_set_init_copy (&local_sctx.limits, &sctx->limits); | ||
| 2210 | if (BE (err != REG_NOERROR, 0)) | ||
| 2211 | goto free_return; | ||
| 2212 | } | ||
| 2213 | local_sctx.last_node = node; | ||
| 2214 | local_sctx.last_str_idx = str_idx; | ||
| 2215 | ok = re_node_set_insert (&local_sctx.limits, enabled_idx); | ||
| 2216 | if (BE (! ok, 0)) | ||
| 2217 | { | ||
| 2218 | err = REG_ESPACE; | ||
| 2219 | goto free_return; | ||
| 2220 | } | ||
| 2221 | cur_state = local_sctx.sifted_states[str_idx]; | ||
| 2222 | err = sift_states_backward (mctx, &local_sctx); | ||
| 2223 | if (BE (err != REG_NOERROR, 0)) | ||
| 2224 | goto free_return; | ||
| 2225 | if (sctx->limited_states != NULL) | ||
| 2226 | { | ||
| 2227 | err = merge_state_array (dfa, sctx->limited_states, | ||
| 2228 | local_sctx.sifted_states, | ||
| 2229 | str_idx + 1); | ||
| 2230 | if (BE (err != REG_NOERROR, 0)) | ||
| 2231 | goto free_return; | ||
| 2232 | } | ||
| 2233 | local_sctx.sifted_states[str_idx] = cur_state; | ||
| 2234 | re_node_set_remove (&local_sctx.limits, enabled_idx); | ||
| 2235 | |||
| 2236 | /* mctx->bkref_ents may have changed, reload the pointer. */ | ||
| 2237 | entry = mctx->bkref_ents + enabled_idx; | ||
| 2238 | } | ||
| 2239 | while (enabled_idx++, entry++->more); | ||
| 2240 | } | ||
| 2241 | err = REG_NOERROR; | ||
| 2242 | free_return: | ||
| 2243 | if (local_sctx.sifted_states != NULL) | ||
| 2244 | { | ||
| 2245 | re_node_set_free (&local_sctx.limits); | ||
| 2246 | } | ||
| 2247 | |||
| 2248 | return err; | ||
| 2249 | } | ||
| 2250 | |||
| 2251 | |||
| 2252 | #ifdef RE_ENABLE_I18N | ||
| 2253 | static int | ||
| 2254 | internal_function | ||
| 2255 | sift_states_iter_mb (const re_match_context_t *mctx, re_sift_context_t *sctx, | ||
| 2256 | Idx node_idx, Idx str_idx, Idx max_str_idx) | ||
| 2257 | { | ||
| 2258 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2259 | int naccepted; | ||
| 2260 | /* Check the node can accept `multi byte'. */ | ||
| 2261 | naccepted = check_node_accept_bytes (dfa, node_idx, &mctx->input, str_idx); | ||
| 2262 | if (naccepted > 0 && str_idx + naccepted <= max_str_idx && | ||
| 2263 | !STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + naccepted], | ||
| 2264 | dfa->nexts[node_idx])) | ||
| 2265 | /* The node can't accept the `multi byte', or the | ||
| 2266 | destination was already thrown away, then the node | ||
| 2267 | could't accept the current input `multi byte'. */ | ||
| 2268 | naccepted = 0; | ||
| 2269 | /* Otherwise, it is sure that the node could accept | ||
| 2270 | `naccepted' bytes input. */ | ||
| 2271 | return naccepted; | ||
| 2272 | } | ||
| 2273 | #endif /* RE_ENABLE_I18N */ | ||
| 2274 | |||
| 2275 | |||
| 2276 | /* Functions for state transition. */ | ||
| 2277 | |||
| 2278 | /* Return the next state to which the current state STATE will transit by | ||
| 2279 | accepting the current input byte, and update STATE_LOG if necessary. | ||
| 2280 | If STATE can accept a multibyte char/collating element/back reference | ||
| 2281 | update the destination of STATE_LOG. */ | ||
| 2282 | |||
| 2283 | static re_dfastate_t * | ||
| 2284 | internal_function | ||
| 2285 | transit_state (reg_errcode_t *err, re_match_context_t *mctx, | ||
| 2286 | re_dfastate_t *state) | ||
| 2287 | { | ||
| 2288 | re_dfastate_t **trtable; | ||
| 2289 | unsigned char ch; | ||
| 2290 | |||
| 2291 | #ifdef RE_ENABLE_I18N | ||
| 2292 | /* If the current state can accept multibyte. */ | ||
| 2293 | if (BE (state->accept_mb, 0)) | ||
| 2294 | { | ||
| 2295 | *err = transit_state_mb (mctx, state); | ||
| 2296 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2297 | return NULL; | ||
| 2298 | } | ||
| 2299 | #endif /* RE_ENABLE_I18N */ | ||
| 2300 | |||
| 2301 | /* Then decide the next state with the single byte. */ | ||
| 2302 | #if 0 | ||
| 2303 | if (0) | ||
| 2304 | /* don't use transition table */ | ||
| 2305 | return transit_state_sb (err, mctx, state); | ||
| 2306 | #endif | ||
| 2307 | |||
| 2308 | /* Use transition table */ | ||
| 2309 | ch = re_string_fetch_byte (&mctx->input); | ||
| 2310 | for (;;) | ||
| 2311 | { | ||
| 2312 | trtable = state->trtable; | ||
| 2313 | if (BE (trtable != NULL, 1)) | ||
| 2314 | return trtable[ch]; | ||
| 2315 | |||
| 2316 | trtable = state->word_trtable; | ||
| 2317 | if (BE (trtable != NULL, 1)) | ||
| 2318 | { | ||
| 2319 | unsigned int context; | ||
| 2320 | context | ||
| 2321 | = re_string_context_at (&mctx->input, | ||
| 2322 | re_string_cur_idx (&mctx->input) - 1, | ||
| 2323 | mctx->eflags); | ||
| 2324 | if (IS_WORD_CONTEXT (context)) | ||
| 2325 | return trtable[ch + SBC_MAX]; | ||
| 2326 | else | ||
| 2327 | return trtable[ch]; | ||
| 2328 | } | ||
| 2329 | |||
| 2330 | if (!build_trtable (mctx->dfa, state)) | ||
| 2331 | { | ||
| 2332 | *err = REG_ESPACE; | ||
| 2333 | return NULL; | ||
| 2334 | } | ||
| 2335 | |||
| 2336 | /* Retry, we now have a transition table. */ | ||
| 2337 | } | ||
| 2338 | } | ||
| 2339 | |||
| 2340 | /* Update the state_log if we need */ | ||
| 2341 | re_dfastate_t * | ||
| 2342 | internal_function | ||
| 2343 | merge_state_with_log (reg_errcode_t *err, re_match_context_t *mctx, | ||
| 2344 | re_dfastate_t *next_state) | ||
| 2345 | { | ||
| 2346 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2347 | Idx cur_idx = re_string_cur_idx (&mctx->input); | ||
| 2348 | |||
| 2349 | if (cur_idx > mctx->state_log_top) | ||
| 2350 | { | ||
| 2351 | mctx->state_log[cur_idx] = next_state; | ||
| 2352 | mctx->state_log_top = cur_idx; | ||
| 2353 | } | ||
| 2354 | else if (mctx->state_log[cur_idx] == 0) | ||
| 2355 | { | ||
| 2356 | mctx->state_log[cur_idx] = next_state; | ||
| 2357 | } | ||
| 2358 | else | ||
| 2359 | { | ||
| 2360 | re_dfastate_t *pstate; | ||
| 2361 | unsigned int context; | ||
| 2362 | re_node_set next_nodes, *log_nodes, *table_nodes = NULL; | ||
| 2363 | /* If (state_log[cur_idx] != 0), it implies that cur_idx is | ||
| 2364 | the destination of a multibyte char/collating element/ | ||
| 2365 | back reference. Then the next state is the union set of | ||
| 2366 | these destinations and the results of the transition table. */ | ||
| 2367 | pstate = mctx->state_log[cur_idx]; | ||
| 2368 | log_nodes = pstate->entrance_nodes; | ||
| 2369 | if (next_state != NULL) | ||
| 2370 | { | ||
| 2371 | table_nodes = next_state->entrance_nodes; | ||
| 2372 | *err = re_node_set_init_union (&next_nodes, table_nodes, | ||
| 2373 | log_nodes); | ||
| 2374 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2375 | return NULL; | ||
| 2376 | } | ||
| 2377 | else | ||
| 2378 | next_nodes = *log_nodes; | ||
| 2379 | /* Note: We already add the nodes of the initial state, | ||
| 2380 | then we don't need to add them here. */ | ||
| 2381 | |||
| 2382 | context = re_string_context_at (&mctx->input, | ||
| 2383 | re_string_cur_idx (&mctx->input) - 1, | ||
| 2384 | mctx->eflags); | ||
| 2385 | next_state = mctx->state_log[cur_idx] | ||
| 2386 | = re_acquire_state_context (err, dfa, &next_nodes, context); | ||
| 2387 | /* We don't need to check errors here, since the return value of | ||
| 2388 | this function is next_state and ERR is already set. */ | ||
| 2389 | |||
| 2390 | if (table_nodes != NULL) | ||
| 2391 | re_node_set_free (&next_nodes); | ||
| 2392 | } | ||
| 2393 | |||
| 2394 | if (BE (dfa->nbackref, 0) && next_state != NULL) | ||
| 2395 | { | ||
| 2396 | /* Check OP_OPEN_SUBEXP in the current state in case that we use them | ||
| 2397 | later. We must check them here, since the back references in the | ||
| 2398 | next state might use them. */ | ||
| 2399 | *err = check_subexp_matching_top (mctx, &next_state->nodes, | ||
| 2400 | cur_idx); | ||
| 2401 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2402 | return NULL; | ||
| 2403 | |||
| 2404 | /* If the next state has back references. */ | ||
| 2405 | if (next_state->has_backref) | ||
| 2406 | { | ||
| 2407 | *err = transit_state_bkref (mctx, &next_state->nodes); | ||
| 2408 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2409 | return NULL; | ||
| 2410 | next_state = mctx->state_log[cur_idx]; | ||
| 2411 | } | ||
| 2412 | } | ||
| 2413 | |||
| 2414 | return next_state; | ||
| 2415 | } | ||
| 2416 | |||
| 2417 | /* Skip bytes in the input that correspond to part of a | ||
| 2418 | multi-byte match, then look in the log for a state | ||
| 2419 | from which to restart matching. */ | ||
| 2420 | static re_dfastate_t * | ||
| 2421 | internal_function | ||
| 2422 | find_recover_state (reg_errcode_t *err, re_match_context_t *mctx) | ||
| 2423 | { | ||
| 2424 | re_dfastate_t *cur_state; | ||
| 2425 | do | ||
| 2426 | { | ||
| 2427 | Idx max = mctx->state_log_top; | ||
| 2428 | Idx cur_str_idx = re_string_cur_idx (&mctx->input); | ||
| 2429 | |||
| 2430 | do | ||
| 2431 | { | ||
| 2432 | if (++cur_str_idx > max) | ||
| 2433 | return NULL; | ||
| 2434 | re_string_skip_bytes (&mctx->input, 1); | ||
| 2435 | } | ||
| 2436 | while (mctx->state_log[cur_str_idx] == NULL); | ||
| 2437 | |||
| 2438 | cur_state = merge_state_with_log (err, mctx, NULL); | ||
| 2439 | } | ||
| 2440 | while (*err == REG_NOERROR && cur_state == NULL); | ||
| 2441 | return cur_state; | ||
| 2442 | } | ||
| 2443 | |||
| 2444 | /* Helper functions for transit_state. */ | ||
| 2445 | |||
| 2446 | /* From the node set CUR_NODES, pick up the nodes whose types are | ||
| 2447 | OP_OPEN_SUBEXP and which have corresponding back references in the regular | ||
| 2448 | expression. And register them to use them later for evaluating the | ||
| 2449 | correspoding back references. */ | ||
| 2450 | |||
| 2451 | static reg_errcode_t | ||
| 2452 | internal_function | ||
| 2453 | check_subexp_matching_top (re_match_context_t *mctx, re_node_set *cur_nodes, | ||
| 2454 | Idx str_idx) | ||
| 2455 | { | ||
| 2456 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2457 | Idx node_idx; | ||
| 2458 | reg_errcode_t err; | ||
| 2459 | |||
| 2460 | /* TODO: This isn't efficient. | ||
| 2461 | Because there might be more than one nodes whose types are | ||
| 2462 | OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all | ||
| 2463 | nodes. | ||
| 2464 | E.g. RE: (a){2} */ | ||
| 2465 | for (node_idx = 0; node_idx < cur_nodes->nelem; ++node_idx) | ||
| 2466 | { | ||
| 2467 | Idx node = cur_nodes->elems[node_idx]; | ||
| 2468 | if (dfa->nodes[node].type == OP_OPEN_SUBEXP | ||
| 2469 | && dfa->nodes[node].opr.idx < BITSET_WORD_BITS | ||
| 2470 | && (dfa->used_bkref_map | ||
| 2471 | & ((bitset_word_t) 1 << dfa->nodes[node].opr.idx))) | ||
| 2472 | { | ||
| 2473 | err = match_ctx_add_subtop (mctx, node, str_idx); | ||
| 2474 | if (BE (err != REG_NOERROR, 0)) | ||
| 2475 | return err; | ||
| 2476 | } | ||
| 2477 | } | ||
| 2478 | return REG_NOERROR; | ||
| 2479 | } | ||
| 2480 | |||
| 2481 | #if 0 | ||
| 2482 | /* Return the next state to which the current state STATE will transit by | ||
| 2483 | accepting the current input byte. */ | ||
| 2484 | |||
| 2485 | static re_dfastate_t * | ||
| 2486 | transit_state_sb (reg_errcode_t *err, re_match_context_t *mctx, | ||
| 2487 | re_dfastate_t *state) | ||
| 2488 | { | ||
| 2489 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2490 | re_node_set next_nodes; | ||
| 2491 | re_dfastate_t *next_state; | ||
| 2492 | Idx node_cnt, cur_str_idx = re_string_cur_idx (&mctx->input); | ||
| 2493 | unsigned int context; | ||
| 2494 | |||
| 2495 | *err = re_node_set_alloc (&next_nodes, state->nodes.nelem + 1); | ||
| 2496 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2497 | return NULL; | ||
| 2498 | for (node_cnt = 0; node_cnt < state->nodes.nelem; ++node_cnt) | ||
| 2499 | { | ||
| 2500 | Idx cur_node = state->nodes.elems[node_cnt]; | ||
| 2501 | if (check_node_accept (mctx, dfa->nodes + cur_node, cur_str_idx)) | ||
| 2502 | { | ||
| 2503 | *err = re_node_set_merge (&next_nodes, | ||
| 2504 | dfa->eclosures + dfa->nexts[cur_node]); | ||
| 2505 | if (BE (*err != REG_NOERROR, 0)) | ||
| 2506 | { | ||
| 2507 | re_node_set_free (&next_nodes); | ||
| 2508 | return NULL; | ||
| 2509 | } | ||
| 2510 | } | ||
| 2511 | } | ||
| 2512 | context = re_string_context_at (&mctx->input, cur_str_idx, mctx->eflags); | ||
| 2513 | next_state = re_acquire_state_context (err, dfa, &next_nodes, context); | ||
| 2514 | /* We don't need to check errors here, since the return value of | ||
| 2515 | this function is next_state and ERR is already set. */ | ||
| 2516 | |||
| 2517 | re_node_set_free (&next_nodes); | ||
| 2518 | re_string_skip_bytes (&mctx->input, 1); | ||
| 2519 | return next_state; | ||
| 2520 | } | ||
| 2521 | #endif | ||
| 2522 | |||
| 2523 | #ifdef RE_ENABLE_I18N | ||
| 2524 | static reg_errcode_t | ||
| 2525 | internal_function | ||
| 2526 | transit_state_mb (re_match_context_t *mctx, re_dfastate_t *pstate) | ||
| 2527 | { | ||
| 2528 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2529 | reg_errcode_t err; | ||
| 2530 | Idx i; | ||
| 2531 | |||
| 2532 | for (i = 0; i < pstate->nodes.nelem; ++i) | ||
| 2533 | { | ||
| 2534 | re_node_set dest_nodes, *new_nodes; | ||
| 2535 | Idx cur_node_idx = pstate->nodes.elems[i]; | ||
| 2536 | int naccepted; | ||
| 2537 | Idx dest_idx; | ||
| 2538 | unsigned int context; | ||
| 2539 | re_dfastate_t *dest_state; | ||
| 2540 | |||
| 2541 | if (!dfa->nodes[cur_node_idx].accept_mb) | ||
| 2542 | continue; | ||
| 2543 | |||
| 2544 | if (dfa->nodes[cur_node_idx].constraint) | ||
| 2545 | { | ||
| 2546 | context = re_string_context_at (&mctx->input, | ||
| 2547 | re_string_cur_idx (&mctx->input), | ||
| 2548 | mctx->eflags); | ||
| 2549 | if (NOT_SATISFY_NEXT_CONSTRAINT (dfa->nodes[cur_node_idx].constraint, | ||
| 2550 | context)) | ||
| 2551 | continue; | ||
| 2552 | } | ||
| 2553 | |||
| 2554 | /* How many bytes the node can accept? */ | ||
| 2555 | naccepted = check_node_accept_bytes (dfa, cur_node_idx, &mctx->input, | ||
| 2556 | re_string_cur_idx (&mctx->input)); | ||
| 2557 | if (naccepted == 0) | ||
| 2558 | continue; | ||
| 2559 | |||
| 2560 | /* The node can accepts `naccepted' bytes. */ | ||
| 2561 | dest_idx = re_string_cur_idx (&mctx->input) + naccepted; | ||
| 2562 | mctx->max_mb_elem_len = ((mctx->max_mb_elem_len < naccepted) ? naccepted | ||
| 2563 | : mctx->max_mb_elem_len); | ||
| 2564 | err = clean_state_log_if_needed (mctx, dest_idx); | ||
| 2565 | if (BE (err != REG_NOERROR, 0)) | ||
| 2566 | return err; | ||
| 2567 | #ifdef DEBUG | ||
| 2568 | assert (dfa->nexts[cur_node_idx] != REG_MISSING); | ||
| 2569 | #endif | ||
| 2570 | new_nodes = dfa->eclosures + dfa->nexts[cur_node_idx]; | ||
| 2571 | |||
| 2572 | dest_state = mctx->state_log[dest_idx]; | ||
| 2573 | if (dest_state == NULL) | ||
| 2574 | dest_nodes = *new_nodes; | ||
| 2575 | else | ||
| 2576 | { | ||
| 2577 | err = re_node_set_init_union (&dest_nodes, | ||
| 2578 | dest_state->entrance_nodes, new_nodes); | ||
| 2579 | if (BE (err != REG_NOERROR, 0)) | ||
| 2580 | return err; | ||
| 2581 | } | ||
| 2582 | context = re_string_context_at (&mctx->input, dest_idx - 1, | ||
| 2583 | mctx->eflags); | ||
| 2584 | mctx->state_log[dest_idx] | ||
| 2585 | = re_acquire_state_context (&err, dfa, &dest_nodes, context); | ||
| 2586 | if (dest_state != NULL) | ||
| 2587 | re_node_set_free (&dest_nodes); | ||
| 2588 | if (BE (mctx->state_log[dest_idx] == NULL && err != REG_NOERROR, 0)) | ||
| 2589 | return err; | ||
| 2590 | } | ||
| 2591 | return REG_NOERROR; | ||
| 2592 | } | ||
| 2593 | #endif /* RE_ENABLE_I18N */ | ||
| 2594 | |||
| 2595 | static reg_errcode_t | ||
| 2596 | internal_function | ||
| 2597 | transit_state_bkref (re_match_context_t *mctx, const re_node_set *nodes) | ||
| 2598 | { | ||
| 2599 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2600 | reg_errcode_t err; | ||
| 2601 | Idx i; | ||
| 2602 | Idx cur_str_idx = re_string_cur_idx (&mctx->input); | ||
| 2603 | |||
| 2604 | for (i = 0; i < nodes->nelem; ++i) | ||
| 2605 | { | ||
| 2606 | Idx dest_str_idx, prev_nelem, bkc_idx; | ||
| 2607 | Idx node_idx = nodes->elems[i]; | ||
| 2608 | unsigned int context; | ||
| 2609 | const re_token_t *node = dfa->nodes + node_idx; | ||
| 2610 | re_node_set *new_dest_nodes; | ||
| 2611 | |||
| 2612 | /* Check whether `node' is a backreference or not. */ | ||
| 2613 | if (node->type != OP_BACK_REF) | ||
| 2614 | continue; | ||
| 2615 | |||
| 2616 | if (node->constraint) | ||
| 2617 | { | ||
| 2618 | context = re_string_context_at (&mctx->input, cur_str_idx, | ||
| 2619 | mctx->eflags); | ||
| 2620 | if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context)) | ||
| 2621 | continue; | ||
| 2622 | } | ||
| 2623 | |||
| 2624 | /* `node' is a backreference. | ||
| 2625 | Check the substring which the substring matched. */ | ||
| 2626 | bkc_idx = mctx->nbkref_ents; | ||
| 2627 | err = get_subexp (mctx, node_idx, cur_str_idx); | ||
| 2628 | if (BE (err != REG_NOERROR, 0)) | ||
| 2629 | goto free_return; | ||
| 2630 | |||
| 2631 | /* And add the epsilon closures (which is `new_dest_nodes') of | ||
| 2632 | the backreference to appropriate state_log. */ | ||
| 2633 | #ifdef DEBUG | ||
| 2634 | assert (dfa->nexts[node_idx] != REG_MISSING); | ||
| 2635 | #endif | ||
| 2636 | for (; bkc_idx < mctx->nbkref_ents; ++bkc_idx) | ||
| 2637 | { | ||
| 2638 | Idx subexp_len; | ||
| 2639 | re_dfastate_t *dest_state; | ||
| 2640 | struct re_backref_cache_entry *bkref_ent; | ||
| 2641 | bkref_ent = mctx->bkref_ents + bkc_idx; | ||
| 2642 | if (bkref_ent->node != node_idx || bkref_ent->str_idx != cur_str_idx) | ||
| 2643 | continue; | ||
| 2644 | subexp_len = bkref_ent->subexp_to - bkref_ent->subexp_from; | ||
| 2645 | new_dest_nodes = (subexp_len == 0 | ||
| 2646 | ? dfa->eclosures + dfa->edests[node_idx].elems[0] | ||
| 2647 | : dfa->eclosures + dfa->nexts[node_idx]); | ||
| 2648 | dest_str_idx = (cur_str_idx + bkref_ent->subexp_to | ||
| 2649 | - bkref_ent->subexp_from); | ||
| 2650 | context = re_string_context_at (&mctx->input, dest_str_idx - 1, | ||
| 2651 | mctx->eflags); | ||
| 2652 | dest_state = mctx->state_log[dest_str_idx]; | ||
| 2653 | prev_nelem = ((mctx->state_log[cur_str_idx] == NULL) ? 0 | ||
| 2654 | : mctx->state_log[cur_str_idx]->nodes.nelem); | ||
| 2655 | /* Add `new_dest_node' to state_log. */ | ||
| 2656 | if (dest_state == NULL) | ||
| 2657 | { | ||
| 2658 | mctx->state_log[dest_str_idx] | ||
| 2659 | = re_acquire_state_context (&err, dfa, new_dest_nodes, | ||
| 2660 | context); | ||
| 2661 | if (BE (mctx->state_log[dest_str_idx] == NULL | ||
| 2662 | && err != REG_NOERROR, 0)) | ||
| 2663 | goto free_return; | ||
| 2664 | } | ||
| 2665 | else | ||
| 2666 | { | ||
| 2667 | re_node_set dest_nodes; | ||
| 2668 | err = re_node_set_init_union (&dest_nodes, | ||
| 2669 | dest_state->entrance_nodes, | ||
| 2670 | new_dest_nodes); | ||
| 2671 | if (BE (err != REG_NOERROR, 0)) | ||
| 2672 | { | ||
| 2673 | re_node_set_free (&dest_nodes); | ||
| 2674 | goto free_return; | ||
| 2675 | } | ||
| 2676 | mctx->state_log[dest_str_idx] | ||
| 2677 | = re_acquire_state_context (&err, dfa, &dest_nodes, context); | ||
| 2678 | re_node_set_free (&dest_nodes); | ||
| 2679 | if (BE (mctx->state_log[dest_str_idx] == NULL | ||
| 2680 | && err != REG_NOERROR, 0)) | ||
| 2681 | goto free_return; | ||
| 2682 | } | ||
| 2683 | /* We need to check recursively if the backreference can epsilon | ||
| 2684 | transit. */ | ||
| 2685 | if (subexp_len == 0 | ||
| 2686 | && mctx->state_log[cur_str_idx]->nodes.nelem > prev_nelem) | ||
| 2687 | { | ||
| 2688 | err = check_subexp_matching_top (mctx, new_dest_nodes, | ||
| 2689 | cur_str_idx); | ||
| 2690 | if (BE (err != REG_NOERROR, 0)) | ||
| 2691 | goto free_return; | ||
| 2692 | err = transit_state_bkref (mctx, new_dest_nodes); | ||
| 2693 | if (BE (err != REG_NOERROR, 0)) | ||
| 2694 | goto free_return; | ||
| 2695 | } | ||
| 2696 | } | ||
| 2697 | } | ||
| 2698 | err = REG_NOERROR; | ||
| 2699 | free_return: | ||
| 2700 | return err; | ||
| 2701 | } | ||
| 2702 | |||
| 2703 | /* Enumerate all the candidates which the backreference BKREF_NODE can match | ||
| 2704 | at BKREF_STR_IDX, and register them by match_ctx_add_entry(). | ||
| 2705 | Note that we might collect inappropriate candidates here. | ||
| 2706 | However, the cost of checking them strictly here is too high, then we | ||
| 2707 | delay these checking for prune_impossible_nodes(). */ | ||
| 2708 | |||
| 2709 | static reg_errcode_t | ||
| 2710 | internal_function | ||
| 2711 | get_subexp (re_match_context_t *mctx, Idx bkref_node, Idx bkref_str_idx) | ||
| 2712 | { | ||
| 2713 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2714 | Idx subexp_num, sub_top_idx; | ||
| 2715 | const char *buf = (const char *) re_string_get_buffer (&mctx->input); | ||
| 2716 | /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */ | ||
| 2717 | Idx cache_idx = search_cur_bkref_entry (mctx, bkref_str_idx); | ||
| 2718 | if (cache_idx != REG_MISSING) | ||
| 2719 | { | ||
| 2720 | const struct re_backref_cache_entry *entry | ||
| 2721 | = mctx->bkref_ents + cache_idx; | ||
| 2722 | do | ||
| 2723 | if (entry->node == bkref_node) | ||
| 2724 | return REG_NOERROR; /* We already checked it. */ | ||
| 2725 | while (entry++->more); | ||
| 2726 | } | ||
| 2727 | |||
| 2728 | subexp_num = dfa->nodes[bkref_node].opr.idx; | ||
| 2729 | |||
| 2730 | /* For each sub expression */ | ||
| 2731 | for (sub_top_idx = 0; sub_top_idx < mctx->nsub_tops; ++sub_top_idx) | ||
| 2732 | { | ||
| 2733 | reg_errcode_t err; | ||
| 2734 | re_sub_match_top_t *sub_top = mctx->sub_tops[sub_top_idx]; | ||
| 2735 | re_sub_match_last_t *sub_last; | ||
| 2736 | Idx sub_last_idx, sl_str, bkref_str_off; | ||
| 2737 | |||
| 2738 | if (dfa->nodes[sub_top->node].opr.idx != subexp_num) | ||
| 2739 | continue; /* It isn't related. */ | ||
| 2740 | |||
| 2741 | sl_str = sub_top->str_idx; | ||
| 2742 | bkref_str_off = bkref_str_idx; | ||
| 2743 | /* At first, check the last node of sub expressions we already | ||
| 2744 | evaluated. */ | ||
| 2745 | for (sub_last_idx = 0; sub_last_idx < sub_top->nlasts; ++sub_last_idx) | ||
| 2746 | { | ||
| 2747 | regoff_t sl_str_diff; | ||
| 2748 | sub_last = sub_top->lasts[sub_last_idx]; | ||
| 2749 | sl_str_diff = sub_last->str_idx - sl_str; | ||
| 2750 | /* The matched string by the sub expression match with the substring | ||
| 2751 | at the back reference? */ | ||
| 2752 | if (sl_str_diff > 0) | ||
| 2753 | { | ||
| 2754 | if (BE (bkref_str_off + sl_str_diff > mctx->input.valid_len, 0)) | ||
| 2755 | { | ||
| 2756 | /* Not enough chars for a successful match. */ | ||
| 2757 | if (bkref_str_off + sl_str_diff > mctx->input.len) | ||
| 2758 | break; | ||
| 2759 | |||
| 2760 | err = clean_state_log_if_needed (mctx, | ||
| 2761 | bkref_str_off | ||
| 2762 | + sl_str_diff); | ||
| 2763 | if (BE (err != REG_NOERROR, 0)) | ||
| 2764 | return err; | ||
| 2765 | buf = (const char *) re_string_get_buffer (&mctx->input); | ||
| 2766 | } | ||
| 2767 | if (memcmp (buf + bkref_str_off, buf + sl_str, sl_str_diff) != 0) | ||
| 2768 | /* We don't need to search this sub expression any more. */ | ||
| 2769 | break; | ||
| 2770 | } | ||
| 2771 | bkref_str_off += sl_str_diff; | ||
| 2772 | sl_str += sl_str_diff; | ||
| 2773 | err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node, | ||
| 2774 | bkref_str_idx); | ||
| 2775 | |||
| 2776 | /* Reload buf, since the preceding call might have reallocated | ||
| 2777 | the buffer. */ | ||
| 2778 | buf = (const char *) re_string_get_buffer (&mctx->input); | ||
| 2779 | |||
| 2780 | if (err == REG_NOMATCH) | ||
| 2781 | continue; | ||
| 2782 | if (BE (err != REG_NOERROR, 0)) | ||
| 2783 | return err; | ||
| 2784 | } | ||
| 2785 | |||
| 2786 | if (sub_last_idx < sub_top->nlasts) | ||
| 2787 | continue; | ||
| 2788 | if (sub_last_idx > 0) | ||
| 2789 | ++sl_str; | ||
| 2790 | /* Then, search for the other last nodes of the sub expression. */ | ||
| 2791 | for (; sl_str <= bkref_str_idx; ++sl_str) | ||
| 2792 | { | ||
| 2793 | Idx cls_node; | ||
| 2794 | regoff_t sl_str_off; | ||
| 2795 | const re_node_set *nodes; | ||
| 2796 | sl_str_off = sl_str - sub_top->str_idx; | ||
| 2797 | /* The matched string by the sub expression match with the substring | ||
| 2798 | at the back reference? */ | ||
| 2799 | if (sl_str_off > 0) | ||
| 2800 | { | ||
| 2801 | if (BE (bkref_str_off >= mctx->input.valid_len, 0)) | ||
| 2802 | { | ||
| 2803 | /* If we are at the end of the input, we cannot match. */ | ||
| 2804 | if (bkref_str_off >= mctx->input.len) | ||
| 2805 | break; | ||
| 2806 | |||
| 2807 | err = extend_buffers (mctx); | ||
| 2808 | if (BE (err != REG_NOERROR, 0)) | ||
| 2809 | return err; | ||
| 2810 | |||
| 2811 | buf = (const char *) re_string_get_buffer (&mctx->input); | ||
| 2812 | } | ||
| 2813 | if (buf [bkref_str_off++] != buf[sl_str - 1]) | ||
| 2814 | break; /* We don't need to search this sub expression | ||
| 2815 | any more. */ | ||
| 2816 | } | ||
| 2817 | if (mctx->state_log[sl_str] == NULL) | ||
| 2818 | continue; | ||
| 2819 | /* Does this state have a ')' of the sub expression? */ | ||
| 2820 | nodes = &mctx->state_log[sl_str]->nodes; | ||
| 2821 | cls_node = find_subexp_node (dfa, nodes, subexp_num, | ||
| 2822 | OP_CLOSE_SUBEXP); | ||
| 2823 | if (cls_node == REG_MISSING) | ||
| 2824 | continue; /* No. */ | ||
| 2825 | if (sub_top->path == NULL) | ||
| 2826 | { | ||
| 2827 | sub_top->path = calloc (sizeof (state_array_t), | ||
| 2828 | sl_str - sub_top->str_idx + 1); | ||
| 2829 | if (sub_top->path == NULL) | ||
| 2830 | return REG_ESPACE; | ||
| 2831 | } | ||
| 2832 | /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node | ||
| 2833 | in the current context? */ | ||
| 2834 | err = check_arrival (mctx, sub_top->path, sub_top->node, | ||
| 2835 | sub_top->str_idx, cls_node, sl_str, | ||
| 2836 | OP_CLOSE_SUBEXP); | ||
| 2837 | if (err == REG_NOMATCH) | ||
| 2838 | continue; | ||
| 2839 | if (BE (err != REG_NOERROR, 0)) | ||
| 2840 | return err; | ||
| 2841 | sub_last = match_ctx_add_sublast (sub_top, cls_node, sl_str); | ||
| 2842 | if (BE (sub_last == NULL, 0)) | ||
| 2843 | return REG_ESPACE; | ||
| 2844 | err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node, | ||
| 2845 | bkref_str_idx); | ||
| 2846 | if (err == REG_NOMATCH) | ||
| 2847 | continue; | ||
| 2848 | } | ||
| 2849 | } | ||
| 2850 | return REG_NOERROR; | ||
| 2851 | } | ||
| 2852 | |||
| 2853 | /* Helper functions for get_subexp(). */ | ||
| 2854 | |||
| 2855 | /* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR. | ||
| 2856 | If it can arrive, register the sub expression expressed with SUB_TOP | ||
| 2857 | and SUB_LAST. */ | ||
| 2858 | |||
| 2859 | static reg_errcode_t | ||
| 2860 | internal_function | ||
| 2861 | get_subexp_sub (re_match_context_t *mctx, const re_sub_match_top_t *sub_top, | ||
| 2862 | re_sub_match_last_t *sub_last, Idx bkref_node, Idx bkref_str) | ||
| 2863 | { | ||
| 2864 | reg_errcode_t err; | ||
| 2865 | Idx to_idx; | ||
| 2866 | /* Can the subexpression arrive the back reference? */ | ||
| 2867 | err = check_arrival (mctx, &sub_last->path, sub_last->node, | ||
| 2868 | sub_last->str_idx, bkref_node, bkref_str, | ||
| 2869 | OP_OPEN_SUBEXP); | ||
| 2870 | if (err != REG_NOERROR) | ||
| 2871 | return err; | ||
| 2872 | err = match_ctx_add_entry (mctx, bkref_node, bkref_str, sub_top->str_idx, | ||
| 2873 | sub_last->str_idx); | ||
| 2874 | if (BE (err != REG_NOERROR, 0)) | ||
| 2875 | return err; | ||
| 2876 | to_idx = bkref_str + sub_last->str_idx - sub_top->str_idx; | ||
| 2877 | return clean_state_log_if_needed (mctx, to_idx); | ||
| 2878 | } | ||
| 2879 | |||
| 2880 | /* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX. | ||
| 2881 | Search '(' if FL_OPEN, or search ')' otherwise. | ||
| 2882 | TODO: This function isn't efficient... | ||
| 2883 | Because there might be more than one nodes whose types are | ||
| 2884 | OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all | ||
| 2885 | nodes. | ||
| 2886 | E.g. RE: (a){2} */ | ||
| 2887 | |||
| 2888 | static Idx | ||
| 2889 | internal_function | ||
| 2890 | find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes, | ||
| 2891 | Idx subexp_idx, int type) | ||
| 2892 | { | ||
| 2893 | Idx cls_idx; | ||
| 2894 | for (cls_idx = 0; cls_idx < nodes->nelem; ++cls_idx) | ||
| 2895 | { | ||
| 2896 | Idx cls_node = nodes->elems[cls_idx]; | ||
| 2897 | const re_token_t *node = dfa->nodes + cls_node; | ||
| 2898 | if (node->type == type | ||
| 2899 | && node->opr.idx == subexp_idx) | ||
| 2900 | return cls_node; | ||
| 2901 | } | ||
| 2902 | return REG_MISSING; | ||
| 2903 | } | ||
| 2904 | |||
| 2905 | /* Check whether the node TOP_NODE at TOP_STR can arrive to the node | ||
| 2906 | LAST_NODE at LAST_STR. We record the path onto PATH since it will be | ||
| 2907 | heavily reused. | ||
| 2908 | Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */ | ||
| 2909 | |||
| 2910 | static reg_errcode_t | ||
| 2911 | internal_function | ||
| 2912 | check_arrival (re_match_context_t *mctx, state_array_t *path, Idx top_node, | ||
| 2913 | Idx top_str, Idx last_node, Idx last_str, int type) | ||
| 2914 | { | ||
| 2915 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 2916 | reg_errcode_t err = REG_NOERROR; | ||
| 2917 | Idx subexp_num, backup_cur_idx, str_idx, null_cnt; | ||
| 2918 | re_dfastate_t *cur_state = NULL; | ||
| 2919 | re_node_set *cur_nodes, next_nodes; | ||
| 2920 | re_dfastate_t **backup_state_log; | ||
| 2921 | unsigned int context; | ||
| 2922 | |||
| 2923 | subexp_num = dfa->nodes[top_node].opr.idx; | ||
| 2924 | /* Extend the buffer if we need. */ | ||
| 2925 | if (BE (path->alloc < last_str + mctx->max_mb_elem_len + 1, 0)) | ||
| 2926 | { | ||
| 2927 | re_dfastate_t **new_array; | ||
| 2928 | Idx old_alloc = path->alloc; | ||
| 2929 | Idx new_alloc = old_alloc + last_str + mctx->max_mb_elem_len + 1; | ||
| 2930 | if (BE (new_alloc < old_alloc, 0) | ||
| 2931 | || BE (SIZE_MAX / sizeof (re_dfastate_t *) < new_alloc, 0)) | ||
| 2932 | return REG_ESPACE; | ||
| 2933 | new_array = re_realloc (path->array, re_dfastate_t *, new_alloc); | ||
| 2934 | if (BE (new_array == NULL, 0)) | ||
| 2935 | return REG_ESPACE; | ||
| 2936 | path->array = new_array; | ||
| 2937 | path->alloc = new_alloc; | ||
| 2938 | memset (new_array + old_alloc, '\0', | ||
| 2939 | sizeof (re_dfastate_t *) * (path->alloc - old_alloc)); | ||
| 2940 | } | ||
| 2941 | |||
| 2942 | str_idx = path->next_idx ? path->next_idx : top_str; | ||
| 2943 | |||
| 2944 | /* Temporary modify MCTX. */ | ||
| 2945 | backup_state_log = mctx->state_log; | ||
| 2946 | backup_cur_idx = mctx->input.cur_idx; | ||
| 2947 | mctx->state_log = path->array; | ||
| 2948 | mctx->input.cur_idx = str_idx; | ||
| 2949 | |||
| 2950 | /* Setup initial node set. */ | ||
| 2951 | context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags); | ||
| 2952 | if (str_idx == top_str) | ||
| 2953 | { | ||
| 2954 | err = re_node_set_init_1 (&next_nodes, top_node); | ||
| 2955 | if (BE (err != REG_NOERROR, 0)) | ||
| 2956 | return err; | ||
| 2957 | err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type); | ||
| 2958 | if (BE (err != REG_NOERROR, 0)) | ||
| 2959 | { | ||
| 2960 | re_node_set_free (&next_nodes); | ||
| 2961 | return err; | ||
| 2962 | } | ||
| 2963 | } | ||
| 2964 | else | ||
| 2965 | { | ||
| 2966 | cur_state = mctx->state_log[str_idx]; | ||
| 2967 | if (cur_state && cur_state->has_backref) | ||
| 2968 | { | ||
| 2969 | err = re_node_set_init_copy (&next_nodes, &cur_state->nodes); | ||
| 2970 | if (BE (err != REG_NOERROR, 0)) | ||
| 2971 | return err; | ||
| 2972 | } | ||
| 2973 | else | ||
| 2974 | re_node_set_init_empty (&next_nodes); | ||
| 2975 | } | ||
| 2976 | if (str_idx == top_str || (cur_state && cur_state->has_backref)) | ||
| 2977 | { | ||
| 2978 | if (next_nodes.nelem) | ||
| 2979 | { | ||
| 2980 | err = expand_bkref_cache (mctx, &next_nodes, str_idx, | ||
| 2981 | subexp_num, type); | ||
| 2982 | if (BE (err != REG_NOERROR, 0)) | ||
| 2983 | { | ||
| 2984 | re_node_set_free (&next_nodes); | ||
| 2985 | return err; | ||
| 2986 | } | ||
| 2987 | } | ||
| 2988 | cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context); | ||
| 2989 | if (BE (cur_state == NULL && err != REG_NOERROR, 0)) | ||
| 2990 | { | ||
| 2991 | re_node_set_free (&next_nodes); | ||
| 2992 | return err; | ||
| 2993 | } | ||
| 2994 | mctx->state_log[str_idx] = cur_state; | ||
| 2995 | } | ||
| 2996 | |||
| 2997 | for (null_cnt = 0; str_idx < last_str && null_cnt <= mctx->max_mb_elem_len;) | ||
| 2998 | { | ||
| 2999 | re_node_set_empty (&next_nodes); | ||
| 3000 | if (mctx->state_log[str_idx + 1]) | ||
| 3001 | { | ||
| 3002 | err = re_node_set_merge (&next_nodes, | ||
| 3003 | &mctx->state_log[str_idx + 1]->nodes); | ||
| 3004 | if (BE (err != REG_NOERROR, 0)) | ||
| 3005 | { | ||
| 3006 | re_node_set_free (&next_nodes); | ||
| 3007 | return err; | ||
| 3008 | } | ||
| 3009 | } | ||
| 3010 | if (cur_state) | ||
| 3011 | { | ||
| 3012 | err = check_arrival_add_next_nodes (mctx, str_idx, | ||
| 3013 | &cur_state->non_eps_nodes, | ||
| 3014 | &next_nodes); | ||
| 3015 | if (BE (err != REG_NOERROR, 0)) | ||
| 3016 | { | ||
| 3017 | re_node_set_free (&next_nodes); | ||
| 3018 | return err; | ||
| 3019 | } | ||
| 3020 | } | ||
| 3021 | ++str_idx; | ||
| 3022 | if (next_nodes.nelem) | ||
| 3023 | { | ||
| 3024 | err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type); | ||
| 3025 | if (BE (err != REG_NOERROR, 0)) | ||
| 3026 | { | ||
| 3027 | re_node_set_free (&next_nodes); | ||
| 3028 | return err; | ||
| 3029 | } | ||
| 3030 | err = expand_bkref_cache (mctx, &next_nodes, str_idx, | ||
| 3031 | subexp_num, type); | ||
| 3032 | if (BE (err != REG_NOERROR, 0)) | ||
| 3033 | { | ||
| 3034 | re_node_set_free (&next_nodes); | ||
| 3035 | return err; | ||
| 3036 | } | ||
| 3037 | } | ||
| 3038 | context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags); | ||
| 3039 | cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context); | ||
| 3040 | if (BE (cur_state == NULL && err != REG_NOERROR, 0)) | ||
| 3041 | { | ||
| 3042 | re_node_set_free (&next_nodes); | ||
| 3043 | return err; | ||
| 3044 | } | ||
| 3045 | mctx->state_log[str_idx] = cur_state; | ||
| 3046 | null_cnt = cur_state == NULL ? null_cnt + 1 : 0; | ||
| 3047 | } | ||
| 3048 | re_node_set_free (&next_nodes); | ||
| 3049 | cur_nodes = (mctx->state_log[last_str] == NULL ? NULL | ||
| 3050 | : &mctx->state_log[last_str]->nodes); | ||
| 3051 | path->next_idx = str_idx; | ||
| 3052 | |||
| 3053 | /* Fix MCTX. */ | ||
| 3054 | mctx->state_log = backup_state_log; | ||
| 3055 | mctx->input.cur_idx = backup_cur_idx; | ||
| 3056 | |||
| 3057 | /* Then check the current node set has the node LAST_NODE. */ | ||
| 3058 | if (cur_nodes != NULL && re_node_set_contains (cur_nodes, last_node)) | ||
| 3059 | return REG_NOERROR; | ||
| 3060 | |||
| 3061 | return REG_NOMATCH; | ||
| 3062 | } | ||
| 3063 | |||
| 3064 | /* Helper functions for check_arrival. */ | ||
| 3065 | |||
| 3066 | /* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them | ||
| 3067 | to NEXT_NODES. | ||
| 3068 | TODO: This function is similar to the functions transit_state*(), | ||
| 3069 | however this function has many additional works. | ||
| 3070 | Can't we unify them? */ | ||
| 3071 | |||
| 3072 | static reg_errcode_t | ||
| 3073 | internal_function | ||
| 3074 | check_arrival_add_next_nodes (re_match_context_t *mctx, Idx str_idx, | ||
| 3075 | re_node_set *cur_nodes, re_node_set *next_nodes) | ||
| 3076 | { | ||
| 3077 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 3078 | bool ok; | ||
| 3079 | Idx cur_idx; | ||
| 3080 | reg_errcode_t err = REG_NOERROR; | ||
| 3081 | re_node_set union_set; | ||
| 3082 | re_node_set_init_empty (&union_set); | ||
| 3083 | for (cur_idx = 0; cur_idx < cur_nodes->nelem; ++cur_idx) | ||
| 3084 | { | ||
| 3085 | int naccepted = 0; | ||
| 3086 | Idx cur_node = cur_nodes->elems[cur_idx]; | ||
| 3087 | #ifdef DEBUG | ||
| 3088 | re_token_type_t type = dfa->nodes[cur_node].type; | ||
| 3089 | assert (!IS_EPSILON_NODE (type)); | ||
| 3090 | #endif | ||
| 3091 | #ifdef RE_ENABLE_I18N | ||
| 3092 | /* If the node may accept `multi byte'. */ | ||
| 3093 | if (dfa->nodes[cur_node].accept_mb) | ||
| 3094 | { | ||
| 3095 | naccepted = check_node_accept_bytes (dfa, cur_node, &mctx->input, | ||
| 3096 | str_idx); | ||
| 3097 | if (naccepted > 1) | ||
| 3098 | { | ||
| 3099 | re_dfastate_t *dest_state; | ||
| 3100 | Idx next_node = dfa->nexts[cur_node]; | ||
| 3101 | Idx next_idx = str_idx + naccepted; | ||
| 3102 | dest_state = mctx->state_log[next_idx]; | ||
| 3103 | re_node_set_empty (&union_set); | ||
| 3104 | if (dest_state) | ||
| 3105 | { | ||
| 3106 | err = re_node_set_merge (&union_set, &dest_state->nodes); | ||
| 3107 | if (BE (err != REG_NOERROR, 0)) | ||
| 3108 | { | ||
| 3109 | re_node_set_free (&union_set); | ||
| 3110 | return err; | ||
| 3111 | } | ||
| 3112 | } | ||
| 3113 | ok = re_node_set_insert (&union_set, next_node); | ||
| 3114 | if (BE (! ok, 0)) | ||
| 3115 | { | ||
| 3116 | re_node_set_free (&union_set); | ||
| 3117 | return REG_ESPACE; | ||
| 3118 | } | ||
| 3119 | mctx->state_log[next_idx] = re_acquire_state (&err, dfa, | ||
| 3120 | &union_set); | ||
| 3121 | if (BE (mctx->state_log[next_idx] == NULL | ||
| 3122 | && err != REG_NOERROR, 0)) | ||
| 3123 | { | ||
| 3124 | re_node_set_free (&union_set); | ||
| 3125 | return err; | ||
| 3126 | } | ||
| 3127 | } | ||
| 3128 | } | ||
| 3129 | #endif /* RE_ENABLE_I18N */ | ||
| 3130 | if (naccepted | ||
| 3131 | || check_node_accept (mctx, dfa->nodes + cur_node, str_idx)) | ||
| 3132 | { | ||
| 3133 | ok = re_node_set_insert (next_nodes, dfa->nexts[cur_node]); | ||
| 3134 | if (BE (! ok, 0)) | ||
| 3135 | { | ||
| 3136 | re_node_set_free (&union_set); | ||
| 3137 | return REG_ESPACE; | ||
| 3138 | } | ||
| 3139 | } | ||
| 3140 | } | ||
| 3141 | re_node_set_free (&union_set); | ||
| 3142 | return REG_NOERROR; | ||
| 3143 | } | ||
| 3144 | |||
| 3145 | /* For all the nodes in CUR_NODES, add the epsilon closures of them to | ||
| 3146 | CUR_NODES, however exclude the nodes which are: | ||
| 3147 | - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN. | ||
| 3148 | - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN. | ||
| 3149 | */ | ||
| 3150 | |||
| 3151 | static reg_errcode_t | ||
| 3152 | internal_function | ||
| 3153 | check_arrival_expand_ecl (const re_dfa_t *dfa, re_node_set *cur_nodes, | ||
| 3154 | Idx ex_subexp, int type) | ||
| 3155 | { | ||
| 3156 | reg_errcode_t err; | ||
| 3157 | Idx idx, outside_node; | ||
| 3158 | re_node_set new_nodes; | ||
| 3159 | #ifdef DEBUG | ||
| 3160 | assert (cur_nodes->nelem); | ||
| 3161 | #endif | ||
| 3162 | err = re_node_set_alloc (&new_nodes, cur_nodes->nelem); | ||
| 3163 | if (BE (err != REG_NOERROR, 0)) | ||
| 3164 | return err; | ||
| 3165 | /* Create a new node set NEW_NODES with the nodes which are epsilon | ||
| 3166 | closures of the node in CUR_NODES. */ | ||
| 3167 | |||
| 3168 | for (idx = 0; idx < cur_nodes->nelem; ++idx) | ||
| 3169 | { | ||
| 3170 | Idx cur_node = cur_nodes->elems[idx]; | ||
| 3171 | const re_node_set *eclosure = dfa->eclosures + cur_node; | ||
| 3172 | outside_node = find_subexp_node (dfa, eclosure, ex_subexp, type); | ||
| 3173 | if (outside_node == REG_MISSING) | ||
| 3174 | { | ||
| 3175 | /* There are no problematic nodes, just merge them. */ | ||
| 3176 | err = re_node_set_merge (&new_nodes, eclosure); | ||
| 3177 | if (BE (err != REG_NOERROR, 0)) | ||
| 3178 | { | ||
| 3179 | re_node_set_free (&new_nodes); | ||
| 3180 | return err; | ||
| 3181 | } | ||
| 3182 | } | ||
| 3183 | else | ||
| 3184 | { | ||
| 3185 | /* There are problematic nodes, re-calculate incrementally. */ | ||
| 3186 | err = check_arrival_expand_ecl_sub (dfa, &new_nodes, cur_node, | ||
| 3187 | ex_subexp, type); | ||
| 3188 | if (BE (err != REG_NOERROR, 0)) | ||
| 3189 | { | ||
| 3190 | re_node_set_free (&new_nodes); | ||
| 3191 | return err; | ||
| 3192 | } | ||
| 3193 | } | ||
| 3194 | } | ||
| 3195 | re_node_set_free (cur_nodes); | ||
| 3196 | *cur_nodes = new_nodes; | ||
| 3197 | return REG_NOERROR; | ||
| 3198 | } | ||
| 3199 | |||
| 3200 | /* Helper function for check_arrival_expand_ecl. | ||
| 3201 | Check incrementally the epsilon closure of TARGET, and if it isn't | ||
| 3202 | problematic append it to DST_NODES. */ | ||
| 3203 | |||
| 3204 | static reg_errcode_t | ||
| 3205 | internal_function | ||
| 3206 | check_arrival_expand_ecl_sub (const re_dfa_t *dfa, re_node_set *dst_nodes, | ||
| 3207 | Idx target, Idx ex_subexp, int type) | ||
| 3208 | { | ||
| 3209 | Idx cur_node; | ||
| 3210 | for (cur_node = target; !re_node_set_contains (dst_nodes, cur_node);) | ||
| 3211 | { | ||
| 3212 | bool ok; | ||
| 3213 | |||
| 3214 | if (dfa->nodes[cur_node].type == type | ||
| 3215 | && dfa->nodes[cur_node].opr.idx == ex_subexp) | ||
| 3216 | { | ||
| 3217 | if (type == OP_CLOSE_SUBEXP) | ||
| 3218 | { | ||
| 3219 | ok = re_node_set_insert (dst_nodes, cur_node); | ||
| 3220 | if (BE (! ok, 0)) | ||
| 3221 | return REG_ESPACE; | ||
| 3222 | } | ||
| 3223 | break; | ||
| 3224 | } | ||
| 3225 | ok = re_node_set_insert (dst_nodes, cur_node); | ||
| 3226 | if (BE (! ok, 0)) | ||
| 3227 | return REG_ESPACE; | ||
| 3228 | if (dfa->edests[cur_node].nelem == 0) | ||
| 3229 | break; | ||
| 3230 | if (dfa->edests[cur_node].nelem == 2) | ||
| 3231 | { | ||
| 3232 | reg_errcode_t err; | ||
| 3233 | err = check_arrival_expand_ecl_sub (dfa, dst_nodes, | ||
| 3234 | dfa->edests[cur_node].elems[1], | ||
| 3235 | ex_subexp, type); | ||
| 3236 | if (BE (err != REG_NOERROR, 0)) | ||
| 3237 | return err; | ||
| 3238 | } | ||
| 3239 | cur_node = dfa->edests[cur_node].elems[0]; | ||
| 3240 | } | ||
| 3241 | return REG_NOERROR; | ||
| 3242 | } | ||
| 3243 | |||
| 3244 | |||
| 3245 | /* For all the back references in the current state, calculate the | ||
| 3246 | destination of the back references by the appropriate entry | ||
| 3247 | in MCTX->BKREF_ENTS. */ | ||
| 3248 | |||
| 3249 | static reg_errcode_t | ||
| 3250 | internal_function | ||
| 3251 | expand_bkref_cache (re_match_context_t *mctx, re_node_set *cur_nodes, | ||
| 3252 | Idx cur_str, Idx subexp_num, int type) | ||
| 3253 | { | ||
| 3254 | const re_dfa_t *const dfa = mctx->dfa; | ||
| 3255 | reg_errcode_t err; | ||
| 3256 | Idx cache_idx_start = search_cur_bkref_entry (mctx, cur_str); | ||
| 3257 | struct re_backref_cache_entry *ent; | ||
| 3258 | |||
| 3259 | if (cache_idx_start == REG_MISSING) | ||
| 3260 | return REG_NOERROR; | ||
| 3261 | |||
| 3262 | restart: | ||
| 3263 | ent = mctx->bkref_ents + cache_idx_start; | ||
| 3264 | do | ||
| 3265 | { | ||
| 3266 | Idx to_idx, next_node; | ||
| 3267 | |||
| 3268 | /* Is this entry ENT is appropriate? */ | ||
| 3269 | if (!re_node_set_contains (cur_nodes, ent->node)) | ||
| 3270 | continue; /* No. */ | ||
| 3271 | |||
| 3272 | to_idx = cur_str + ent->subexp_to - ent->subexp_from; | ||
| 3273 | /* Calculate the destination of the back reference, and append it | ||
| 3274 | to MCTX->STATE_LOG. */ | ||
| 3275 | if (to_idx == cur_str) | ||
| 3276 | { | ||
| 3277 | /* The backreference did epsilon transit, we must re-check all the | ||
| 3278 | node in the current state. */ | ||
| 3279 | re_node_set new_dests; | ||
| 3280 | reg_errcode_t err2, err3; | ||
| 3281 | next_node = dfa->edests[ent->node].elems[0]; | ||
| 3282 | if (re_node_set_contains (cur_nodes, next_node)) | ||
| 3283 | continue; | ||
| 3284 | err = re_node_set_init_1 (&new_dests, next_node); | ||
| 3285 | err2 = check_arrival_expand_ecl (dfa, &new_dests, subexp_num, type); | ||
| 3286 | err3 = re_node_set_merge (cur_nodes, &new_dests); | ||
| 3287 | re_node_set_free (&new_dests); | ||
| 3288 | if (BE (err != REG_NOERROR || err2 != REG_NOERROR | ||
| 3289 | || err3 != REG_NOERROR, 0)) | ||
| 3290 | { | ||
| 3291 | err = (err != REG_NOERROR ? err | ||
| 3292 | : (err2 != REG_NOERROR ? err2 : err3)); | ||
| 3293 | return err; | ||
| 3294 | } | ||
| 3295 | /* TODO: It is still inefficient... */ | ||
| 3296 | goto restart; | ||
| 3297 | } | ||
| 3298 | else | ||
| 3299 | { | ||
| 3300 | re_node_set union_set; | ||
| 3301 | next_node = dfa->nexts[ent->node]; | ||
| 3302 | if (mctx->state_log[to_idx]) | ||
| 3303 | { | ||
| 3304 | bool ok; | ||
| 3305 | if (re_node_set_contains (&mctx->state_log[to_idx]->nodes, | ||
| 3306 | next_node)) | ||
| 3307 | continue; | ||
| 3308 | err = re_node_set_init_copy (&union_set, | ||
| 3309 | &mctx->state_log[to_idx]->nodes); | ||
| 3310 | ok = re_node_set_insert (&union_set, next_node); | ||
| 3311 | if (BE (err != REG_NOERROR || ! ok, 0)) | ||
| 3312 | { | ||
| 3313 | re_node_set_free (&union_set); | ||
| 3314 | err = err != REG_NOERROR ? err : REG_ESPACE; | ||
| 3315 | return err; | ||
| 3316 | } | ||
| 3317 | } | ||
| 3318 | else | ||
| 3319 | { | ||
| 3320 | err = re_node_set_init_1 (&union_set, next_node); | ||
| 3321 | if (BE (err != REG_NOERROR, 0)) | ||
| 3322 | return err; | ||
| 3323 | } | ||
| 3324 | mctx->state_log[to_idx] = re_acquire_state (&err, dfa, &union_set); | ||
| 3325 | re_node_set_free (&union_set); | ||
| 3326 | if (BE (mctx->state_log[to_idx] == NULL | ||
| 3327 | && err != REG_NOERROR, 0)) | ||
| 3328 | return err; | ||
| 3329 | } | ||
| 3330 | } | ||
| 3331 | while (ent++->more); | ||
| 3332 | return REG_NOERROR; | ||
| 3333 | } | ||
| 3334 | |||
| 3335 | /* Build transition table for the state. | ||
| 3336 | Return true if successful. */ | ||
| 3337 | |||
| 3338 | static bool | ||
| 3339 | internal_function | ||
| 3340 | build_trtable (const re_dfa_t *dfa, re_dfastate_t *state) | ||
| 3341 | { | ||
| 3342 | reg_errcode_t err; | ||
| 3343 | Idx i, j; | ||
| 3344 | int ch; | ||
| 3345 | bool need_word_trtable = false; | ||
| 3346 | bitset_word_t elem, mask; | ||
| 3347 | bool dests_node_malloced = false; | ||
| 3348 | bool dest_states_malloced = false; | ||
| 3349 | Idx ndests; /* Number of the destination states from `state'. */ | ||
| 3350 | re_dfastate_t **trtable; | ||
| 3351 | re_dfastate_t **dest_states = NULL, **dest_states_word, **dest_states_nl; | ||
| 3352 | re_node_set follows, *dests_node; | ||
| 3353 | bitset_t *dests_ch; | ||
| 3354 | bitset_t acceptable; | ||
| 3355 | |||
| 3356 | struct dests_alloc | ||
| 3357 | { | ||
| 3358 | re_node_set dests_node[SBC_MAX]; | ||
| 3359 | bitset_t dests_ch[SBC_MAX]; | ||
| 3360 | } *dests_alloc; | ||
| 3361 | |||
| 3362 | /* We build DFA states which corresponds to the destination nodes | ||
| 3363 | from `state'. `dests_node[i]' represents the nodes which i-th | ||
| 3364 | destination state contains, and `dests_ch[i]' represents the | ||
| 3365 | characters which i-th destination state accepts. */ | ||
| 3366 | if (__libc_use_alloca (sizeof (struct dests_alloc))) | ||
| 3367 | dests_alloc = (struct dests_alloc *) alloca (sizeof (struct dests_alloc)); | ||
| 3368 | else | ||
| 3369 | { | ||
| 3370 | dests_alloc = re_malloc (struct dests_alloc, 1); | ||
| 3371 | if (BE (dests_alloc == NULL, 0)) | ||
| 3372 | return false; | ||
| 3373 | dests_node_malloced = true; | ||
| 3374 | } | ||
| 3375 | dests_node = dests_alloc->dests_node; | ||
| 3376 | dests_ch = dests_alloc->dests_ch; | ||
| 3377 | |||
| 3378 | /* Initialize transiton table. */ | ||
| 3379 | state->word_trtable = state->trtable = NULL; | ||
| 3380 | |||
| 3381 | /* At first, group all nodes belonging to `state' into several | ||
| 3382 | destinations. */ | ||
| 3383 | ndests = group_nodes_into_DFAstates (dfa, state, dests_node, dests_ch); | ||
| 3384 | if (BE (! REG_VALID_NONZERO_INDEX (ndests), 0)) | ||
| 3385 | { | ||
| 3386 | if (dests_node_malloced) | ||
| 3387 | free (dests_alloc); | ||
| 3388 | if (ndests == 0) | ||
| 3389 | { | ||
| 3390 | state->trtable = (re_dfastate_t **) | ||
| 3391 | calloc (sizeof (re_dfastate_t *), SBC_MAX); | ||
| 3392 | return true; | ||
| 3393 | } | ||
| 3394 | return false; | ||
| 3395 | } | ||
| 3396 | |||
| 3397 | err = re_node_set_alloc (&follows, ndests + 1); | ||
| 3398 | if (BE (err != REG_NOERROR, 0)) | ||
| 3399 | goto out_free; | ||
| 3400 | |||
| 3401 | /* Avoid arithmetic overflow in size calculation. */ | ||
| 3402 | if (BE ((((SIZE_MAX - (sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX) | ||
| 3403 | / (3 * sizeof (re_dfastate_t *))) | ||
| 3404 | < ndests), | ||
| 3405 | 0)) | ||
| 3406 | goto out_free; | ||
| 3407 | |||
| 3408 | if (__libc_use_alloca ((sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX | ||
| 3409 | + ndests * 3 * sizeof (re_dfastate_t *))) | ||
| 3410 | dest_states = (re_dfastate_t **) | ||
| 3411 | alloca (ndests * 3 * sizeof (re_dfastate_t *)); | ||
| 3412 | else | ||
| 3413 | { | ||
| 3414 | dest_states = (re_dfastate_t **) | ||
| 3415 | malloc (ndests * 3 * sizeof (re_dfastate_t *)); | ||
| 3416 | if (BE (dest_states == NULL, 0)) | ||
| 3417 | { | ||
| 3418 | out_free: | ||
| 3419 | if (dest_states_malloced) | ||
| 3420 | free (dest_states); | ||
| 3421 | re_node_set_free (&follows); | ||
| 3422 | for (i = 0; i < ndests; ++i) | ||
| 3423 | re_node_set_free (dests_node + i); | ||
| 3424 | if (dests_node_malloced) | ||
| 3425 | free (dests_alloc); | ||
| 3426 | return false; | ||
| 3427 | } | ||
| 3428 | dest_states_malloced = true; | ||
| 3429 | } | ||
| 3430 | dest_states_word = dest_states + ndests; | ||
| 3431 | dest_states_nl = dest_states_word + ndests; | ||
| 3432 | bitset_empty (acceptable); | ||
| 3433 | |||
| 3434 | /* Then build the states for all destinations. */ | ||
| 3435 | for (i = 0; i < ndests; ++i) | ||
| 3436 | { | ||
| 3437 | Idx next_node; | ||
| 3438 | re_node_set_empty (&follows); | ||
| 3439 | /* Merge the follows of this destination states. */ | ||
| 3440 | for (j = 0; j < dests_node[i].nelem; ++j) | ||
| 3441 | { | ||
| 3442 | next_node = dfa->nexts[dests_node[i].elems[j]]; | ||
| 3443 | if (next_node != REG_MISSING) | ||
| 3444 | { | ||
| 3445 | err = re_node_set_merge (&follows, dfa->eclosures + next_node); | ||
| 3446 | if (BE (err != REG_NOERROR, 0)) | ||
| 3447 | goto out_free; | ||
| 3448 | } | ||
| 3449 | } | ||
| 3450 | dest_states[i] = re_acquire_state_context (&err, dfa, &follows, 0); | ||
| 3451 | if (BE (dest_states[i] == NULL && err != REG_NOERROR, 0)) | ||
| 3452 | goto out_free; | ||
| 3453 | /* If the new state has context constraint, | ||
| 3454 | build appropriate states for these contexts. */ | ||
| 3455 | if (dest_states[i]->has_constraint) | ||
| 3456 | { | ||
| 3457 | dest_states_word[i] = re_acquire_state_context (&err, dfa, &follows, | ||
| 3458 | CONTEXT_WORD); | ||
| 3459 | if (BE (dest_states_word[i] == NULL && err != REG_NOERROR, 0)) | ||
| 3460 | goto out_free; | ||
| 3461 | |||
| 3462 | if (dest_states[i] != dest_states_word[i] && dfa->mb_cur_max > 1) | ||
| 3463 | need_word_trtable = true; | ||
| 3464 | |||
| 3465 | dest_states_nl[i] = re_acquire_state_context (&err, dfa, &follows, | ||
| 3466 | CONTEXT_NEWLINE); | ||
| 3467 | if (BE (dest_states_nl[i] == NULL && err != REG_NOERROR, 0)) | ||
| 3468 | goto out_free; | ||
| 3469 | } | ||
| 3470 | else | ||
| 3471 | { | ||
| 3472 | dest_states_word[i] = dest_states[i]; | ||
| 3473 | dest_states_nl[i] = dest_states[i]; | ||
| 3474 | } | ||
| 3475 | bitset_merge (acceptable, dests_ch[i]); | ||
| 3476 | } | ||
| 3477 | |||
| 3478 | if (!BE (need_word_trtable, 0)) | ||
| 3479 | { | ||
| 3480 | /* We don't care about whether the following character is a word | ||
| 3481 | character, or we are in a single-byte character set so we can | ||
| 3482 | discern by looking at the character code: allocate a | ||
| 3483 | 256-entry transition table. */ | ||
| 3484 | trtable = state->trtable = | ||
| 3485 | (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), SBC_MAX); | ||
| 3486 | if (BE (trtable == NULL, 0)) | ||
| 3487 | goto out_free; | ||
| 3488 | |||
| 3489 | /* For all characters ch...: */ | ||
| 3490 | for (i = 0; i < BITSET_WORDS; ++i) | ||
| 3491 | for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1; | ||
| 3492 | elem; | ||
| 3493 | mask <<= 1, elem >>= 1, ++ch) | ||
| 3494 | if (BE (elem & 1, 0)) | ||
| 3495 | { | ||
| 3496 | /* There must be exactly one destination which accepts | ||
| 3497 | character ch. See group_nodes_into_DFAstates. */ | ||
| 3498 | for (j = 0; (dests_ch[j][i] & mask) == 0; ++j) | ||
| 3499 | ; | ||
| 3500 | |||
| 3501 | /* j-th destination accepts the word character ch. */ | ||
| 3502 | if (dfa->word_char[i] & mask) | ||
| 3503 | trtable[ch] = dest_states_word[j]; | ||
| 3504 | else | ||
| 3505 | trtable[ch] = dest_states[j]; | ||
| 3506 | } | ||
| 3507 | } | ||
| 3508 | else | ||
| 3509 | { | ||
| 3510 | /* We care about whether the following character is a word | ||
| 3511 | character, and we are in a multi-byte character set: discern | ||
| 3512 | by looking at the character code: build two 256-entry | ||
| 3513 | transition tables, one starting at trtable[0] and one | ||
| 3514 | starting at trtable[SBC_MAX]. */ | ||
| 3515 | trtable = state->word_trtable = | ||
| 3516 | (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), 2 * SBC_MAX); | ||
| 3517 | if (BE (trtable == NULL, 0)) | ||
| 3518 | goto out_free; | ||
| 3519 | |||
| 3520 | /* For all characters ch...: */ | ||
| 3521 | for (i = 0; i < BITSET_WORDS; ++i) | ||
| 3522 | for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1; | ||
| 3523 | elem; | ||
| 3524 | mask <<= 1, elem >>= 1, ++ch) | ||
| 3525 | if (BE (elem & 1, 0)) | ||
| 3526 | { | ||
| 3527 | /* There must be exactly one destination which accepts | ||
| 3528 | character ch. See group_nodes_into_DFAstates. */ | ||
| 3529 | for (j = 0; (dests_ch[j][i] & mask) == 0; ++j) | ||
| 3530 | ; | ||
| 3531 | |||
| 3532 | /* j-th destination accepts the word character ch. */ | ||
| 3533 | trtable[ch] = dest_states[j]; | ||
| 3534 | trtable[ch + SBC_MAX] = dest_states_word[j]; | ||
| 3535 | } | ||
| 3536 | } | ||
| 3537 | |||
| 3538 | /* new line */ | ||
| 3539 | if (bitset_contain (acceptable, NEWLINE_CHAR)) | ||
| 3540 | { | ||
| 3541 | /* The current state accepts newline character. */ | ||
| 3542 | for (j = 0; j < ndests; ++j) | ||
| 3543 | if (bitset_contain (dests_ch[j], NEWLINE_CHAR)) | ||
| 3544 | { | ||
| 3545 | /* k-th destination accepts newline character. */ | ||
| 3546 | trtable[NEWLINE_CHAR] = dest_states_nl[j]; | ||
| 3547 | if (need_word_trtable) | ||
| 3548 | trtable[NEWLINE_CHAR + SBC_MAX] = dest_states_nl[j]; | ||
| 3549 | /* There must be only one destination which accepts | ||
| 3550 | newline. See group_nodes_into_DFAstates. */ | ||
| 3551 | break; | ||
| 3552 | } | ||
| 3553 | } | ||
| 3554 | |||
| 3555 | if (dest_states_malloced) | ||
| 3556 | free (dest_states); | ||
| 3557 | |||
| 3558 | re_node_set_free (&follows); | ||
| 3559 | for (i = 0; i < ndests; ++i) | ||
| 3560 | re_node_set_free (dests_node + i); | ||
| 3561 | |||
| 3562 | if (dests_node_malloced) | ||
| 3563 | free (dests_alloc); | ||
| 3564 | |||
| 3565 | return true; | ||
| 3566 | } | ||
| 3567 | |||
| 3568 | /* Group all nodes belonging to STATE into several destinations. | ||
| 3569 | Then for all destinations, set the nodes belonging to the destination | ||
| 3570 | to DESTS_NODE[i] and set the characters accepted by the destination | ||
| 3571 | to DEST_CH[i]. This function return the number of destinations. */ | ||
| 3572 | |||
| 3573 | static Idx | ||
| 3574 | internal_function | ||
| 3575 | group_nodes_into_DFAstates (const re_dfa_t *dfa, const re_dfastate_t *state, | ||
| 3576 | re_node_set *dests_node, bitset_t *dests_ch) | ||
| 3577 | { | ||
| 3578 | reg_errcode_t err; | ||
| 3579 | bool ok; | ||
| 3580 | Idx i, j, k; | ||
| 3581 | Idx ndests; /* Number of the destinations from `state'. */ | ||
| 3582 | bitset_t accepts; /* Characters a node can accept. */ | ||
| 3583 | const re_node_set *cur_nodes = &state->nodes; | ||
| 3584 | bitset_empty (accepts); | ||
| 3585 | ndests = 0; | ||
| 3586 | |||
| 3587 | /* For all the nodes belonging to `state', */ | ||
| 3588 | for (i = 0; i < cur_nodes->nelem; ++i) | ||
| 3589 | { | ||
| 3590 | re_token_t *node = &dfa->nodes[cur_nodes->elems[i]]; | ||
| 3591 | re_token_type_t type = node->type; | ||
| 3592 | unsigned int constraint = node->constraint; | ||
| 3593 | |||
| 3594 | /* Enumerate all single byte character this node can accept. */ | ||
| 3595 | if (type == CHARACTER) | ||
| 3596 | bitset_set (accepts, node->opr.c); | ||
| 3597 | else if (type == SIMPLE_BRACKET) | ||
| 3598 | { | ||
| 3599 | bitset_merge (accepts, node->opr.sbcset); | ||
| 3600 | } | ||
| 3601 | else if (type == OP_PERIOD) | ||
| 3602 | { | ||
| 3603 | #ifdef RE_ENABLE_I18N | ||
| 3604 | if (dfa->mb_cur_max > 1) | ||
| 3605 | bitset_merge (accepts, dfa->sb_char); | ||
| 3606 | else | ||
| 3607 | #endif | ||
| 3608 | bitset_set_all (accepts); | ||
| 3609 | if (!(dfa->syntax & RE_DOT_NEWLINE)) | ||
| 3610 | bitset_clear (accepts, '\n'); | ||
| 3611 | if (dfa->syntax & RE_DOT_NOT_NULL) | ||
| 3612 | bitset_clear (accepts, '\0'); | ||
| 3613 | } | ||
| 3614 | #ifdef RE_ENABLE_I18N | ||
| 3615 | else if (type == OP_UTF8_PERIOD) | ||
| 3616 | { | ||
| 3617 | if (ASCII_CHARS % BITSET_WORD_BITS == 0) | ||
| 3618 | memset (accepts, -1, ASCII_CHARS / CHAR_BIT); | ||
| 3619 | else | ||
| 3620 | bitset_merge (accepts, utf8_sb_map); | ||
| 3621 | if (!(dfa->syntax & RE_DOT_NEWLINE)) | ||
| 3622 | bitset_clear (accepts, '\n'); | ||
| 3623 | if (dfa->syntax & RE_DOT_NOT_NULL) | ||
| 3624 | bitset_clear (accepts, '\0'); | ||
| 3625 | } | ||
| 3626 | #endif | ||
| 3627 | else | ||
| 3628 | continue; | ||
| 3629 | |||
| 3630 | /* Check the `accepts' and sift the characters which are not | ||
| 3631 | match it the context. */ | ||
| 3632 | if (constraint) | ||
| 3633 | { | ||
| 3634 | if (constraint & NEXT_NEWLINE_CONSTRAINT) | ||
| 3635 | { | ||
| 3636 | bool accepts_newline = bitset_contain (accepts, NEWLINE_CHAR); | ||
| 3637 | bitset_empty (accepts); | ||
| 3638 | if (accepts_newline) | ||
| 3639 | bitset_set (accepts, NEWLINE_CHAR); | ||
| 3640 | else | ||
| 3641 | continue; | ||
| 3642 | } | ||
| 3643 | if (constraint & NEXT_ENDBUF_CONSTRAINT) | ||
| 3644 | { | ||
| 3645 | bitset_empty (accepts); | ||
| 3646 | continue; | ||
| 3647 | } | ||
| 3648 | |||
| 3649 | if (constraint & NEXT_WORD_CONSTRAINT) | ||
| 3650 | { | ||
| 3651 | bitset_word_t any_set = 0; | ||
| 3652 | if (type == CHARACTER && !node->word_char) | ||
| 3653 | { | ||
| 3654 | bitset_empty (accepts); | ||
| 3655 | continue; | ||
| 3656 | } | ||
| 3657 | #ifdef RE_ENABLE_I18N | ||
| 3658 | if (dfa->mb_cur_max > 1) | ||
| 3659 | for (j = 0; j < BITSET_WORDS; ++j) | ||
| 3660 | any_set |= (accepts[j] &= (dfa->word_char[j] | ~dfa->sb_char[j])); | ||
| 3661 | else | ||
| 3662 | #endif | ||
| 3663 | for (j = 0; j < BITSET_WORDS; ++j) | ||
| 3664 | any_set |= (accepts[j] &= dfa->word_char[j]); | ||
| 3665 | if (!any_set) | ||
| 3666 | continue; | ||
| 3667 | } | ||
| 3668 | if (constraint & NEXT_NOTWORD_CONSTRAINT) | ||
| 3669 | { | ||
| 3670 | bitset_word_t any_set = 0; | ||
| 3671 | if (type == CHARACTER && node->word_char) | ||
| 3672 | { | ||
| 3673 | bitset_empty (accepts); | ||
| 3674 | continue; | ||
| 3675 | } | ||
| 3676 | #ifdef RE_ENABLE_I18N | ||
| 3677 | if (dfa->mb_cur_max > 1) | ||
| 3678 | for (j = 0; j < BITSET_WORDS; ++j) | ||
| 3679 | any_set |= (accepts[j] &= ~(dfa->word_char[j] & dfa->sb_char[j])); | ||
| 3680 | else | ||
| 3681 | #endif | ||
| 3682 | for (j = 0; j < BITSET_WORDS; ++j) | ||
| 3683 | any_set |= (accepts[j] &= ~dfa->word_char[j]); | ||
| 3684 | if (!any_set) | ||
| 3685 | continue; | ||
| 3686 | } | ||
| 3687 | } | ||
| 3688 | |||
| 3689 | /* Then divide `accepts' into DFA states, or create a new | ||
| 3690 | state. Above, we make sure that accepts is not empty. */ | ||
| 3691 | for (j = 0; j < ndests; ++j) | ||
| 3692 | { | ||
| 3693 | bitset_t intersec; /* Intersection sets, see below. */ | ||
| 3694 | bitset_t remains; | ||
| 3695 | /* Flags, see below. */ | ||
| 3696 | bitset_word_t has_intersec, not_subset, not_consumed; | ||
| 3697 | |||
| 3698 | /* Optimization, skip if this state doesn't accept the character. */ | ||
| 3699 | if (type == CHARACTER && !bitset_contain (dests_ch[j], node->opr.c)) | ||
| 3700 | continue; | ||
| 3701 | |||
| 3702 | /* Enumerate the intersection set of this state and `accepts'. */ | ||
| 3703 | has_intersec = 0; | ||
| 3704 | for (k = 0; k < BITSET_WORDS; ++k) | ||
| 3705 | has_intersec |= intersec[k] = accepts[k] & dests_ch[j][k]; | ||
| 3706 | /* And skip if the intersection set is empty. */ | ||
| 3707 | if (!has_intersec) | ||
| 3708 | continue; | ||
| 3709 | |||
| 3710 | /* Then check if this state is a subset of `accepts'. */ | ||
| 3711 | not_subset = not_consumed = 0; | ||
| 3712 | for (k = 0; k < BITSET_WORDS; ++k) | ||
| 3713 | { | ||
| 3714 | not_subset |= remains[k] = ~accepts[k] & dests_ch[j][k]; | ||
| 3715 | not_consumed |= accepts[k] = accepts[k] & ~dests_ch[j][k]; | ||
| 3716 | } | ||
| 3717 | |||
| 3718 | /* If this state isn't a subset of `accepts', create a | ||
| 3719 | new group state, which has the `remains'. */ | ||
| 3720 | if (not_subset) | ||
| 3721 | { | ||
| 3722 | bitset_copy (dests_ch[ndests], remains); | ||
| 3723 | bitset_copy (dests_ch[j], intersec); | ||
| 3724 | err = re_node_set_init_copy (dests_node + ndests, &dests_node[j]); | ||
| 3725 | if (BE (err != REG_NOERROR, 0)) | ||
| 3726 | goto error_return; | ||
| 3727 | ++ndests; | ||
| 3728 | } | ||
| 3729 | |||
| 3730 | /* Put the position in the current group. */ | ||
| 3731 | ok = re_node_set_insert (&dests_node[j], cur_nodes->elems[i]); | ||
| 3732 | if (BE (! ok, 0)) | ||
| 3733 | goto error_return; | ||
| 3734 | |||
| 3735 | /* If all characters are consumed, go to next node. */ | ||
| 3736 | if (!not_consumed) | ||
| 3737 | break; | ||
| 3738 | } | ||
| 3739 | /* Some characters remain, create a new group. */ | ||
| 3740 | if (j == ndests) | ||
| 3741 | { | ||
| 3742 | bitset_copy (dests_ch[ndests], accepts); | ||
| 3743 | err = re_node_set_init_1 (dests_node + ndests, cur_nodes->elems[i]); | ||
| 3744 | if (BE (err != REG_NOERROR, 0)) | ||
| 3745 | goto error_return; | ||
| 3746 | ++ndests; | ||
| 3747 | bitset_empty (accepts); | ||
| 3748 | } | ||
| 3749 | } | ||
| 3750 | return ndests; | ||
| 3751 | error_return: | ||
| 3752 | for (j = 0; j < ndests; ++j) | ||
| 3753 | re_node_set_free (dests_node + j); | ||
| 3754 | return REG_MISSING; | ||
| 3755 | } | ||
| 3756 | |||
| 3757 | #ifdef RE_ENABLE_I18N | ||
| 3758 | /* Check how many bytes the node `dfa->nodes[node_idx]' accepts. | ||
| 3759 | Return the number of the bytes the node accepts. | ||
| 3760 | STR_IDX is the current index of the input string. | ||
| 3761 | |||
| 3762 | This function handles the nodes which can accept one character, or | ||
| 3763 | one collating element like '.', '[a-z]', opposite to the other nodes | ||
| 3764 | can only accept one byte. */ | ||
| 3765 | |||
| 3766 | static int | ||
| 3767 | internal_function | ||
| 3768 | check_node_accept_bytes (const re_dfa_t *dfa, Idx node_idx, | ||
| 3769 | const re_string_t *input, Idx str_idx) | ||
| 3770 | { | ||
| 3771 | const re_token_t *node = dfa->nodes + node_idx; | ||
| 3772 | int char_len, elem_len; | ||
| 3773 | Idx i; | ||
| 3774 | |||
| 3775 | if (BE (node->type == OP_UTF8_PERIOD, 0)) | ||
| 3776 | { | ||
| 3777 | unsigned char c = re_string_byte_at (input, str_idx), d; | ||
| 3778 | if (BE (c < 0xc2, 1)) | ||
| 3779 | return 0; | ||
| 3780 | |||
| 3781 | if (str_idx + 2 > input->len) | ||
| 3782 | return 0; | ||
| 3783 | |||
| 3784 | d = re_string_byte_at (input, str_idx + 1); | ||
| 3785 | if (c < 0xe0) | ||
| 3786 | return (d < 0x80 || d > 0xbf) ? 0 : 2; | ||
| 3787 | else if (c < 0xf0) | ||
| 3788 | { | ||
| 3789 | char_len = 3; | ||
| 3790 | if (c == 0xe0 && d < 0xa0) | ||
| 3791 | return 0; | ||
| 3792 | } | ||
| 3793 | else if (c < 0xf8) | ||
| 3794 | { | ||
| 3795 | char_len = 4; | ||
| 3796 | if (c == 0xf0 && d < 0x90) | ||
| 3797 | return 0; | ||
| 3798 | } | ||
| 3799 | else if (c < 0xfc) | ||
| 3800 | { | ||
| 3801 | char_len = 5; | ||
| 3802 | if (c == 0xf8 && d < 0x88) | ||
| 3803 | return 0; | ||
| 3804 | } | ||
| 3805 | else if (c < 0xfe) | ||
| 3806 | { | ||
| 3807 | char_len = 6; | ||
| 3808 | if (c == 0xfc && d < 0x84) | ||
| 3809 | return 0; | ||
| 3810 | } | ||
| 3811 | else | ||
| 3812 | return 0; | ||
| 3813 | |||
| 3814 | if (str_idx + char_len > input->len) | ||
| 3815 | return 0; | ||
| 3816 | |||
| 3817 | for (i = 1; i < char_len; ++i) | ||
| 3818 | { | ||
| 3819 | d = re_string_byte_at (input, str_idx + i); | ||
| 3820 | if (d < 0x80 || d > 0xbf) | ||
| 3821 | return 0; | ||
| 3822 | } | ||
| 3823 | return char_len; | ||
| 3824 | } | ||
| 3825 | |||
| 3826 | char_len = re_string_char_size_at (input, str_idx); | ||
| 3827 | if (node->type == OP_PERIOD) | ||
| 3828 | { | ||
| 3829 | if (char_len <= 1) | ||
| 3830 | return 0; | ||
| 3831 | /* FIXME: I don't think this if is needed, as both '\n' | ||
| 3832 | and '\0' are char_len == 1. */ | ||
| 3833 | /* '.' accepts any one character except the following two cases. */ | ||
| 3834 | if ((!(dfa->syntax & RE_DOT_NEWLINE) && | ||
| 3835 | re_string_byte_at (input, str_idx) == '\n') || | ||
| 3836 | ((dfa->syntax & RE_DOT_NOT_NULL) && | ||
| 3837 | re_string_byte_at (input, str_idx) == '\0')) | ||
| 3838 | return 0; | ||
| 3839 | return char_len; | ||
| 3840 | } | ||
| 3841 | |||
| 3842 | elem_len = re_string_elem_size_at (input, str_idx); | ||
| 3843 | if ((elem_len <= 1 && char_len <= 1) || char_len == 0) | ||
| 3844 | return 0; | ||
| 3845 | |||
| 3846 | if (node->type == COMPLEX_BRACKET) | ||
| 3847 | { | ||
| 3848 | const re_charset_t *cset = node->opr.mbcset; | ||
| 3849 | # ifdef _LIBC | ||
| 3850 | const unsigned char *pin | ||
| 3851 | = ((const unsigned char *) re_string_get_buffer (input) + str_idx); | ||
| 3852 | Idx j; | ||
| 3853 | uint32_t nrules; | ||
| 3854 | # endif /* _LIBC */ | ||
| 3855 | int match_len = 0; | ||
| 3856 | wchar_t wc = ((cset->nranges || cset->nchar_classes || cset->nmbchars) | ||
| 3857 | ? re_string_wchar_at (input, str_idx) : 0); | ||
| 3858 | |||
| 3859 | /* match with multibyte character? */ | ||
| 3860 | for (i = 0; i < cset->nmbchars; ++i) | ||
| 3861 | if (wc == cset->mbchars[i]) | ||
| 3862 | { | ||
| 3863 | match_len = char_len; | ||
| 3864 | goto check_node_accept_bytes_match; | ||
| 3865 | } | ||
| 3866 | /* match with character_class? */ | ||
| 3867 | for (i = 0; i < cset->nchar_classes; ++i) | ||
| 3868 | { | ||
| 3869 | wctype_t wt = cset->char_classes[i]; | ||
| 3870 | if (__iswctype (wc, wt)) | ||
| 3871 | { | ||
| 3872 | match_len = char_len; | ||
| 3873 | goto check_node_accept_bytes_match; | ||
| 3874 | } | ||
| 3875 | } | ||
| 3876 | |||
| 3877 | # ifdef _LIBC | ||
| 3878 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | ||
| 3879 | if (nrules != 0) | ||
| 3880 | { | ||
| 3881 | unsigned int in_collseq = 0; | ||
| 3882 | const int32_t *table, *indirect; | ||
| 3883 | const unsigned char *weights, *extra; | ||
| 3884 | const char *collseqwc; | ||
| 3885 | int32_t idx; | ||
| 3886 | /* This #include defines a local function! */ | ||
| 3887 | # include <locale/weight.h> | ||
| 3888 | |||
| 3889 | /* match with collating_symbol? */ | ||
| 3890 | if (cset->ncoll_syms) | ||
| 3891 | extra = (const unsigned char *) | ||
| 3892 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | ||
| 3893 | for (i = 0; i < cset->ncoll_syms; ++i) | ||
| 3894 | { | ||
| 3895 | const unsigned char *coll_sym = extra + cset->coll_syms[i]; | ||
| 3896 | /* Compare the length of input collating element and | ||
| 3897 | the length of current collating element. */ | ||
| 3898 | if (*coll_sym != elem_len) | ||
| 3899 | continue; | ||
| 3900 | /* Compare each bytes. */ | ||
| 3901 | for (j = 0; j < *coll_sym; j++) | ||
| 3902 | if (pin[j] != coll_sym[1 + j]) | ||
| 3903 | break; | ||
| 3904 | if (j == *coll_sym) | ||
| 3905 | { | ||
| 3906 | /* Match if every bytes is equal. */ | ||
| 3907 | match_len = j; | ||
| 3908 | goto check_node_accept_bytes_match; | ||
| 3909 | } | ||
| 3910 | } | ||
| 3911 | |||
| 3912 | if (cset->nranges) | ||
| 3913 | { | ||
| 3914 | if (elem_len <= char_len) | ||
| 3915 | { | ||
| 3916 | collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC); | ||
| 3917 | in_collseq = __collseq_table_lookup (collseqwc, wc); | ||
| 3918 | } | ||
| 3919 | else | ||
| 3920 | in_collseq = find_collation_sequence_value (pin, elem_len); | ||
| 3921 | } | ||
| 3922 | /* match with range expression? */ | ||
| 3923 | for (i = 0; i < cset->nranges; ++i) | ||
| 3924 | if (cset->range_starts[i] <= in_collseq | ||
| 3925 | && in_collseq <= cset->range_ends[i]) | ||
| 3926 | { | ||
| 3927 | match_len = elem_len; | ||
| 3928 | goto check_node_accept_bytes_match; | ||
| 3929 | } | ||
| 3930 | |||
| 3931 | /* match with equivalence_class? */ | ||
| 3932 | if (cset->nequiv_classes) | ||
| 3933 | { | ||
| 3934 | const unsigned char *cp = pin; | ||
| 3935 | table = (const int32_t *) | ||
| 3936 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); | ||
| 3937 | weights = (const unsigned char *) | ||
| 3938 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); | ||
| 3939 | extra = (const unsigned char *) | ||
| 3940 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); | ||
| 3941 | indirect = (const int32_t *) | ||
| 3942 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); | ||
| 3943 | idx = findidx (&cp); | ||
| 3944 | if (idx > 0) | ||
| 3945 | for (i = 0; i < cset->nequiv_classes; ++i) | ||
| 3946 | { | ||
| 3947 | int32_t equiv_class_idx = cset->equiv_classes[i]; | ||
| 3948 | size_t weight_len = weights[idx]; | ||
| 3949 | if (weight_len == weights[equiv_class_idx]) | ||
| 3950 | { | ||
| 3951 | Idx cnt = 0; | ||
| 3952 | while (cnt <= weight_len | ||
| 3953 | && (weights[equiv_class_idx + 1 + cnt] | ||
| 3954 | == weights[idx + 1 + cnt])) | ||
| 3955 | ++cnt; | ||
| 3956 | if (cnt > weight_len) | ||
| 3957 | { | ||
| 3958 | match_len = elem_len; | ||
| 3959 | goto check_node_accept_bytes_match; | ||
| 3960 | } | ||
| 3961 | } | ||
| 3962 | } | ||
| 3963 | } | ||
| 3964 | } | ||
| 3965 | else | ||
| 3966 | # endif /* _LIBC */ | ||
| 3967 | { | ||
| 3968 | /* match with range expression? */ | ||
| 3969 | #if __GNUC__ >= 2 && ! (__STDC_VERSION__ < 199901L && __STRICT_ANSI__) | ||
| 3970 | wchar_t cmp_buf[] = {L'\0', L'\0', wc, L'\0', L'\0', L'\0'}; | ||
| 3971 | #else | ||
| 3972 | wchar_t cmp_buf[] = {L'\0', L'\0', L'\0', L'\0', L'\0', L'\0'}; | ||
| 3973 | cmp_buf[2] = wc; | ||
| 3974 | #endif | ||
| 3975 | for (i = 0; i < cset->nranges; ++i) | ||
| 3976 | { | ||
| 3977 | cmp_buf[0] = cset->range_starts[i]; | ||
| 3978 | cmp_buf[4] = cset->range_ends[i]; | ||
| 3979 | if (wcscoll (cmp_buf, cmp_buf + 2) <= 0 | ||
| 3980 | && wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0) | ||
| 3981 | { | ||
| 3982 | match_len = char_len; | ||
| 3983 | goto check_node_accept_bytes_match; | ||
| 3984 | } | ||
| 3985 | } | ||
| 3986 | } | ||
| 3987 | check_node_accept_bytes_match: | ||
| 3988 | if (!cset->non_match) | ||
| 3989 | return match_len; | ||
| 3990 | else | ||
| 3991 | { | ||
| 3992 | if (match_len > 0) | ||
| 3993 | return 0; | ||
| 3994 | else | ||
| 3995 | return (elem_len > char_len) ? elem_len : char_len; | ||
| 3996 | } | ||
| 3997 | } | ||
| 3998 | return 0; | ||
| 3999 | } | ||
| 4000 | |||
| 4001 | # ifdef _LIBC | ||
| 4002 | static unsigned int | ||
| 4003 | internal_function | ||
| 4004 | find_collation_sequence_value (const unsigned char *mbs, size_t mbs_len) | ||
| 4005 | { | ||
| 4006 | uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | ||
| 4007 | if (nrules == 0) | ||
| 4008 | { | ||
| 4009 | if (mbs_len == 1) | ||
| 4010 | { | ||
| 4011 | /* No valid character. Match it as a single byte character. */ | ||
| 4012 | const unsigned char *collseq = (const unsigned char *) | ||
| 4013 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB); | ||
| 4014 | return collseq[mbs[0]]; | ||
| 4015 | } | ||
| 4016 | return UINT_MAX; | ||
| 4017 | } | ||
| 4018 | else | ||
| 4019 | { | ||
| 4020 | int32_t idx; | ||
| 4021 | const unsigned char *extra = (const unsigned char *) | ||
| 4022 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | ||
| 4023 | int32_t extrasize = (const unsigned char *) | ||
| 4024 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB + 1) - extra; | ||
| 4025 | |||
| 4026 | for (idx = 0; idx < extrasize;) | ||
| 4027 | { | ||
| 4028 | int mbs_cnt; | ||
| 4029 | bool found = false; | ||
| 4030 | int32_t elem_mbs_len; | ||
| 4031 | /* Skip the name of collating element name. */ | ||
| 4032 | idx = idx + extra[idx] + 1; | ||
| 4033 | elem_mbs_len = extra[idx++]; | ||
| 4034 | if (mbs_len == elem_mbs_len) | ||
| 4035 | { | ||
| 4036 | for (mbs_cnt = 0; mbs_cnt < elem_mbs_len; ++mbs_cnt) | ||
| 4037 | if (extra[idx + mbs_cnt] != mbs[mbs_cnt]) | ||
| 4038 | break; | ||
| 4039 | if (mbs_cnt == elem_mbs_len) | ||
| 4040 | /* Found the entry. */ | ||
| 4041 | found = true; | ||
| 4042 | } | ||
| 4043 | /* Skip the byte sequence of the collating element. */ | ||
| 4044 | idx += elem_mbs_len; | ||
| 4045 | /* Adjust for the alignment. */ | ||
| 4046 | idx = (idx + 3) & ~3; | ||
| 4047 | /* Skip the collation sequence value. */ | ||
| 4048 | idx += sizeof (uint32_t); | ||
| 4049 | /* Skip the wide char sequence of the collating element. */ | ||
| 4050 | idx = idx + sizeof (uint32_t) * (extra[idx] + 1); | ||
| 4051 | /* If we found the entry, return the sequence value. */ | ||
| 4052 | if (found) | ||
| 4053 | return *(uint32_t *) (extra + idx); | ||
| 4054 | /* Skip the collation sequence value. */ | ||
| 4055 | idx += sizeof (uint32_t); | ||
| 4056 | } | ||
| 4057 | return UINT_MAX; | ||
| 4058 | } | ||
| 4059 | } | ||
| 4060 | # endif /* _LIBC */ | ||
| 4061 | #endif /* RE_ENABLE_I18N */ | ||
| 4062 | |||
| 4063 | /* Check whether the node accepts the byte which is IDX-th | ||
| 4064 | byte of the INPUT. */ | ||
| 4065 | |||
| 4066 | static bool | ||
| 4067 | internal_function | ||
| 4068 | check_node_accept (const re_match_context_t *mctx, const re_token_t *node, | ||
| 4069 | Idx idx) | ||
| 4070 | { | ||
| 4071 | unsigned char ch; | ||
| 4072 | ch = re_string_byte_at (&mctx->input, idx); | ||
| 4073 | switch (node->type) | ||
| 4074 | { | ||
| 4075 | case CHARACTER: | ||
| 4076 | if (node->opr.c != ch) | ||
| 4077 | return false; | ||
| 4078 | break; | ||
| 4079 | |||
| 4080 | case SIMPLE_BRACKET: | ||
| 4081 | if (!bitset_contain (node->opr.sbcset, ch)) | ||
| 4082 | return false; | ||
| 4083 | break; | ||
| 4084 | |||
| 4085 | #ifdef RE_ENABLE_I18N | ||
| 4086 | case OP_UTF8_PERIOD: | ||
| 4087 | if (ch >= ASCII_CHARS) | ||
| 4088 | return false; | ||
| 4089 | /* FALLTHROUGH */ | ||
| 4090 | #endif | ||
| 4091 | case OP_PERIOD: | ||
| 4092 | if ((ch == '\n' && !(mctx->dfa->syntax & RE_DOT_NEWLINE)) | ||
| 4093 | || (ch == '\0' && (mctx->dfa->syntax & RE_DOT_NOT_NULL))) | ||
| 4094 | return false; | ||
| 4095 | break; | ||
| 4096 | |||
| 4097 | default: | ||
| 4098 | return false; | ||
| 4099 | } | ||
| 4100 | |||
| 4101 | if (node->constraint) | ||
| 4102 | { | ||
| 4103 | /* The node has constraints. Check whether the current context | ||
| 4104 | satisfies the constraints. */ | ||
| 4105 | unsigned int context = re_string_context_at (&mctx->input, idx, | ||
| 4106 | mctx->eflags); | ||
| 4107 | if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context)) | ||
| 4108 | return false; | ||
| 4109 | } | ||
| 4110 | |||
| 4111 | return true; | ||
| 4112 | } | ||
| 4113 | |||
| 4114 | /* Extend the buffers, if the buffers have run out. */ | ||
| 4115 | |||
| 4116 | static reg_errcode_t | ||
| 4117 | internal_function | ||
| 4118 | extend_buffers (re_match_context_t *mctx) | ||
| 4119 | { | ||
| 4120 | reg_errcode_t ret; | ||
| 4121 | re_string_t *pstr = &mctx->input; | ||
| 4122 | |||
| 4123 | /* Avoid overflow. */ | ||
| 4124 | if (BE (SIZE_MAX / 2 / sizeof (re_dfastate_t *) <= pstr->bufs_len, 0)) | ||
| 4125 | return REG_ESPACE; | ||
| 4126 | |||
| 4127 | /* Double the lengthes of the buffers. */ | ||
| 4128 | ret = re_string_realloc_buffers (pstr, pstr->bufs_len * 2); | ||
| 4129 | if (BE (ret != REG_NOERROR, 0)) | ||
| 4130 | return ret; | ||
| 4131 | |||
| 4132 | if (mctx->state_log != NULL) | ||
| 4133 | { | ||
| 4134 | /* And double the length of state_log. */ | ||
| 4135 | /* XXX We have no indication of the size of this buffer. If this | ||
| 4136 | allocation fail we have no indication that the state_log array | ||
| 4137 | does not have the right size. */ | ||
| 4138 | re_dfastate_t **new_array = re_realloc (mctx->state_log, re_dfastate_t *, | ||
| 4139 | pstr->bufs_len + 1); | ||
| 4140 | if (BE (new_array == NULL, 0)) | ||
| 4141 | return REG_ESPACE; | ||
| 4142 | mctx->state_log = new_array; | ||
| 4143 | } | ||
| 4144 | |||
| 4145 | /* Then reconstruct the buffers. */ | ||
| 4146 | if (pstr->icase) | ||
| 4147 | { | ||
| 4148 | #ifdef RE_ENABLE_I18N | ||
| 4149 | if (pstr->mb_cur_max > 1) | ||
| 4150 | { | ||
| 4151 | ret = build_wcs_upper_buffer (pstr); | ||
| 4152 | if (BE (ret != REG_NOERROR, 0)) | ||
| 4153 | return ret; | ||
| 4154 | } | ||
| 4155 | else | ||
| 4156 | #endif /* RE_ENABLE_I18N */ | ||
| 4157 | build_upper_buffer (pstr); | ||
| 4158 | } | ||
| 4159 | else | ||
| 4160 | { | ||
| 4161 | #ifdef RE_ENABLE_I18N | ||
| 4162 | if (pstr->mb_cur_max > 1) | ||
| 4163 | build_wcs_buffer (pstr); | ||
| 4164 | else | ||
| 4165 | #endif /* RE_ENABLE_I18N */ | ||
| 4166 | { | ||
| 4167 | if (pstr->trans != NULL) | ||
| 4168 | re_string_translate_buffer (pstr); | ||
| 4169 | } | ||
| 4170 | } | ||
| 4171 | return REG_NOERROR; | ||
| 4172 | } | ||
| 4173 | |||
| 4174 | |||
| 4175 | /* Functions for matching context. */ | ||
| 4176 | |||
| 4177 | /* Initialize MCTX. */ | ||
| 4178 | |||
| 4179 | static reg_errcode_t | ||
| 4180 | internal_function | ||
| 4181 | match_ctx_init (re_match_context_t *mctx, int eflags, Idx n) | ||
| 4182 | { | ||
| 4183 | mctx->eflags = eflags; | ||
| 4184 | mctx->match_last = REG_MISSING; | ||
| 4185 | if (n > 0) | ||
| 4186 | { | ||
| 4187 | /* Avoid overflow. */ | ||
| 4188 | size_t max_object_size = | ||
| 4189 | MAX (sizeof (struct re_backref_cache_entry), | ||
| 4190 | sizeof (re_sub_match_top_t *)); | ||
| 4191 | if (BE (SIZE_MAX / max_object_size < n, 0)) | ||
| 4192 | return REG_ESPACE; | ||
| 4193 | |||
| 4194 | mctx->bkref_ents = re_malloc (struct re_backref_cache_entry, n); | ||
| 4195 | mctx->sub_tops = re_malloc (re_sub_match_top_t *, n); | ||
| 4196 | if (BE (mctx->bkref_ents == NULL || mctx->sub_tops == NULL, 0)) | ||
| 4197 | return REG_ESPACE; | ||
| 4198 | } | ||
| 4199 | /* Already zero-ed by the caller. | ||
| 4200 | else | ||
| 4201 | mctx->bkref_ents = NULL; | ||
| 4202 | mctx->nbkref_ents = 0; | ||
| 4203 | mctx->nsub_tops = 0; */ | ||
| 4204 | mctx->abkref_ents = n; | ||
| 4205 | mctx->max_mb_elem_len = 1; | ||
| 4206 | mctx->asub_tops = n; | ||
| 4207 | return REG_NOERROR; | ||
| 4208 | } | ||
| 4209 | |||
| 4210 | /* Clean the entries which depend on the current input in MCTX. | ||
| 4211 | This function must be invoked when the matcher changes the start index | ||
| 4212 | of the input, or changes the input string. */ | ||
| 4213 | |||
| 4214 | static void | ||
| 4215 | internal_function | ||
| 4216 | match_ctx_clean (re_match_context_t *mctx) | ||
| 4217 | { | ||
| 4218 | Idx st_idx; | ||
| 4219 | for (st_idx = 0; st_idx < mctx->nsub_tops; ++st_idx) | ||
| 4220 | { | ||
| 4221 | Idx sl_idx; | ||
| 4222 | re_sub_match_top_t *top = mctx->sub_tops[st_idx]; | ||
| 4223 | for (sl_idx = 0; sl_idx < top->nlasts; ++sl_idx) | ||
| 4224 | { | ||
| 4225 | re_sub_match_last_t *last = top->lasts[sl_idx]; | ||
| 4226 | re_free (last->path.array); | ||
| 4227 | re_free (last); | ||
| 4228 | } | ||
| 4229 | re_free (top->lasts); | ||
| 4230 | if (top->path) | ||
| 4231 | { | ||
| 4232 | re_free (top->path->array); | ||
| 4233 | re_free (top->path); | ||
| 4234 | } | ||
| 4235 | free (top); | ||
| 4236 | } | ||
| 4237 | |||
| 4238 | mctx->nsub_tops = 0; | ||
| 4239 | mctx->nbkref_ents = 0; | ||
| 4240 | } | ||
| 4241 | |||
| 4242 | /* Free all the memory associated with MCTX. */ | ||
| 4243 | |||
| 4244 | static void | ||
| 4245 | internal_function | ||
| 4246 | match_ctx_free (re_match_context_t *mctx) | ||
| 4247 | { | ||
| 4248 | /* First, free all the memory associated with MCTX->SUB_TOPS. */ | ||
| 4249 | match_ctx_clean (mctx); | ||
| 4250 | re_free (mctx->sub_tops); | ||
| 4251 | re_free (mctx->bkref_ents); | ||
| 4252 | } | ||
| 4253 | |||
| 4254 | /* Add a new backreference entry to MCTX. | ||
| 4255 | Note that we assume that caller never call this function with duplicate | ||
| 4256 | entry, and call with STR_IDX which isn't smaller than any existing entry. | ||
| 4257 | */ | ||
| 4258 | |||
| 4259 | static reg_errcode_t | ||
| 4260 | internal_function | ||
| 4261 | match_ctx_add_entry (re_match_context_t *mctx, Idx node, Idx str_idx, Idx from, | ||
| 4262 | Idx to) | ||
| 4263 | { | ||
| 4264 | if (mctx->nbkref_ents >= mctx->abkref_ents) | ||
| 4265 | { | ||
| 4266 | struct re_backref_cache_entry* new_entry; | ||
| 4267 | new_entry = re_realloc (mctx->bkref_ents, struct re_backref_cache_entry, | ||
| 4268 | mctx->abkref_ents * 2); | ||
| 4269 | if (BE (new_entry == NULL, 0)) | ||
| 4270 | { | ||
| 4271 | re_free (mctx->bkref_ents); | ||
| 4272 | return REG_ESPACE; | ||
| 4273 | } | ||
| 4274 | mctx->bkref_ents = new_entry; | ||
| 4275 | memset (mctx->bkref_ents + mctx->nbkref_ents, '\0', | ||
| 4276 | sizeof (struct re_backref_cache_entry) * mctx->abkref_ents); | ||
| 4277 | mctx->abkref_ents *= 2; | ||
| 4278 | } | ||
| 4279 | if (mctx->nbkref_ents > 0 | ||
| 4280 | && mctx->bkref_ents[mctx->nbkref_ents - 1].str_idx == str_idx) | ||
| 4281 | mctx->bkref_ents[mctx->nbkref_ents - 1].more = 1; | ||
| 4282 | |||
| 4283 | mctx->bkref_ents[mctx->nbkref_ents].node = node; | ||
| 4284 | mctx->bkref_ents[mctx->nbkref_ents].str_idx = str_idx; | ||
| 4285 | mctx->bkref_ents[mctx->nbkref_ents].subexp_from = from; | ||
| 4286 | mctx->bkref_ents[mctx->nbkref_ents].subexp_to = to; | ||
| 4287 | |||
| 4288 | /* This is a cache that saves negative results of check_dst_limits_calc_pos. | ||
| 4289 | If bit N is clear, means that this entry won't epsilon-transition to | ||
| 4290 | an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If | ||
| 4291 | it is set, check_dst_limits_calc_pos_1 will recurse and try to find one | ||
| 4292 | such node. | ||
| 4293 | |||
| 4294 | A backreference does not epsilon-transition unless it is empty, so set | ||
| 4295 | to all zeros if FROM != TO. */ | ||
| 4296 | mctx->bkref_ents[mctx->nbkref_ents].eps_reachable_subexps_map | ||
| 4297 | = (from == to ? -1 : 0); | ||
| 4298 | |||
| 4299 | mctx->bkref_ents[mctx->nbkref_ents++].more = 0; | ||
| 4300 | if (mctx->max_mb_elem_len < to - from) | ||
| 4301 | mctx->max_mb_elem_len = to - from; | ||
| 4302 | return REG_NOERROR; | ||
| 4303 | } | ||
| 4304 | |||
| 4305 | /* Return the first entry with the same str_idx, or REG_MISSING if none is | ||
| 4306 | found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */ | ||
| 4307 | |||
| 4308 | static Idx | ||
| 4309 | internal_function | ||
| 4310 | search_cur_bkref_entry (const re_match_context_t *mctx, Idx str_idx) | ||
| 4311 | { | ||
| 4312 | Idx left, right, mid, last; | ||
| 4313 | last = right = mctx->nbkref_ents; | ||
| 4314 | for (left = 0; left < right;) | ||
| 4315 | { | ||
| 4316 | mid = (left + right) / 2; | ||
| 4317 | if (mctx->bkref_ents[mid].str_idx < str_idx) | ||
| 4318 | left = mid + 1; | ||
| 4319 | else | ||
| 4320 | right = mid; | ||
| 4321 | } | ||
| 4322 | if (left < last && mctx->bkref_ents[left].str_idx == str_idx) | ||
| 4323 | return left; | ||
| 4324 | else | ||
| 4325 | return REG_MISSING; | ||
| 4326 | } | ||
| 4327 | |||
| 4328 | /* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches | ||
| 4329 | at STR_IDX. */ | ||
| 4330 | |||
| 4331 | static reg_errcode_t | ||
| 4332 | internal_function | ||
| 4333 | match_ctx_add_subtop (re_match_context_t *mctx, Idx node, Idx str_idx) | ||
| 4334 | { | ||
| 4335 | #ifdef DEBUG | ||
| 4336 | assert (mctx->sub_tops != NULL); | ||
| 4337 | assert (mctx->asub_tops > 0); | ||
| 4338 | #endif | ||
| 4339 | if (BE (mctx->nsub_tops == mctx->asub_tops, 0)) | ||
| 4340 | { | ||
| 4341 | Idx new_asub_tops = mctx->asub_tops * 2; | ||
| 4342 | re_sub_match_top_t **new_array = re_realloc (mctx->sub_tops, | ||
| 4343 | re_sub_match_top_t *, | ||
| 4344 | new_asub_tops); | ||
| 4345 | if (BE (new_array == NULL, 0)) | ||
| 4346 | return REG_ESPACE; | ||
| 4347 | mctx->sub_tops = new_array; | ||
| 4348 | mctx->asub_tops = new_asub_tops; | ||
| 4349 | } | ||
| 4350 | mctx->sub_tops[mctx->nsub_tops] = calloc (1, sizeof (re_sub_match_top_t)); | ||
| 4351 | if (BE (mctx->sub_tops[mctx->nsub_tops] == NULL, 0)) | ||
| 4352 | return REG_ESPACE; | ||
| 4353 | mctx->sub_tops[mctx->nsub_tops]->node = node; | ||
| 4354 | mctx->sub_tops[mctx->nsub_tops++]->str_idx = str_idx; | ||
| 4355 | return REG_NOERROR; | ||
| 4356 | } | ||
| 4357 | |||
| 4358 | /* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches | ||
| 4359 | at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */ | ||
| 4360 | |||
| 4361 | static re_sub_match_last_t * | ||
| 4362 | internal_function | ||
| 4363 | match_ctx_add_sublast (re_sub_match_top_t *subtop, Idx node, Idx str_idx) | ||
| 4364 | { | ||
| 4365 | re_sub_match_last_t *new_entry; | ||
| 4366 | if (BE (subtop->nlasts == subtop->alasts, 0)) | ||
| 4367 | { | ||
| 4368 | Idx new_alasts = 2 * subtop->alasts + 1; | ||
| 4369 | re_sub_match_last_t **new_array = re_realloc (subtop->lasts, | ||
| 4370 | re_sub_match_last_t *, | ||
| 4371 | new_alasts); | ||
| 4372 | if (BE (new_array == NULL, 0)) | ||
| 4373 | return NULL; | ||
| 4374 | subtop->lasts = new_array; | ||
| 4375 | subtop->alasts = new_alasts; | ||
| 4376 | } | ||
| 4377 | new_entry = calloc (1, sizeof (re_sub_match_last_t)); | ||
| 4378 | if (BE (new_entry != NULL, 1)) | ||
| 4379 | { | ||
| 4380 | subtop->lasts[subtop->nlasts] = new_entry; | ||
| 4381 | new_entry->node = node; | ||
| 4382 | new_entry->str_idx = str_idx; | ||
| 4383 | ++subtop->nlasts; | ||
| 4384 | } | ||
| 4385 | return new_entry; | ||
| 4386 | } | ||
| 4387 | |||
| 4388 | static void | ||
| 4389 | internal_function | ||
| 4390 | sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts, | ||
| 4391 | re_dfastate_t **limited_sts, Idx last_node, Idx last_str_idx) | ||
| 4392 | { | ||
| 4393 | sctx->sifted_states = sifted_sts; | ||
| 4394 | sctx->limited_states = limited_sts; | ||
| 4395 | sctx->last_node = last_node; | ||
| 4396 | sctx->last_str_idx = last_str_idx; | ||
| 4397 | re_node_set_init_empty (&sctx->limits); | ||
| 4398 | } | ||
